diff options
Diffstat (limited to 'fs/btrfs/backref.c')
| -rw-r--r-- | fs/btrfs/backref.c | 3340 |
1 files changed, 2330 insertions, 1010 deletions
diff --git a/fs/btrfs/backref.c b/fs/btrfs/backref.c index f723c11bb763..78da47a3d00e 100644 --- a/fs/btrfs/backref.c +++ b/fs/btrfs/backref.c @@ -1,23 +1,11 @@ +// SPDX-License-Identifier: GPL-2.0 /* * Copyright (C) 2011 STRATO. All rights reserved. - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public - * License v2 as published by the Free Software Foundation. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * General Public License for more details. - * - * You should have received a copy of the GNU General Public - * License along with this program; if not, write to the - * Free Software Foundation, Inc., 59 Temple Place - Suite 330, - * Boston, MA 021110-1307, USA. */ #include <linux/mm.h> #include <linux/rbtree.h> +#include <trace/events/btrfs.h> #include "ctree.h" #include "disk-io.h" #include "backref.h" @@ -25,310 +13,79 @@ #include "transaction.h" #include "delayed-ref.h" #include "locking.h" - -enum merge_mode { - MERGE_IDENTICAL_KEYS = 1, - MERGE_IDENTICAL_PARENTS, -}; - -/* Just an arbitrary number so we can be sure this happened */ -#define BACKREF_FOUND_SHARED 6 +#include "misc.h" +#include "tree-mod-log.h" +#include "fs.h" +#include "accessors.h" +#include "extent-tree.h" +#include "relocation.h" +#include "tree-checker.h" + +/* Just arbitrary numbers so we can be sure one of these happened. */ +#define BACKREF_FOUND_SHARED 6 +#define BACKREF_FOUND_NOT_SHARED 7 struct extent_inode_elem { u64 inum; u64 offset; + u64 num_bytes; struct extent_inode_elem *next; }; -/* - * ref_root is used as the root of the ref tree that hold a collection - * of unique references. - */ -struct ref_root { - struct rb_root rb_root; - - /* - * The unique_refs represents the number of ref_nodes with a positive - * count stored in the tree. Even if a ref_node (the count is greater - * than one) is added, the unique_refs will only increase by one. - */ - unsigned int unique_refs; -}; - -/* ref_node is used to store a unique reference to the ref tree. */ -struct ref_node { - struct rb_node rb_node; - - /* For NORMAL_REF, otherwise all these fields should be set to 0 */ - u64 root_id; - u64 object_id; - u64 offset; - - /* For SHARED_REF, otherwise parent field should be set to 0 */ - u64 parent; - - /* Ref to the ref_mod of btrfs_delayed_ref_node */ - int ref_mod; -}; - -/* Dynamically allocate and initialize a ref_root */ -static struct ref_root *ref_root_alloc(void) -{ - struct ref_root *ref_tree; - - ref_tree = kmalloc(sizeof(*ref_tree), GFP_NOFS); - if (!ref_tree) - return NULL; - - ref_tree->rb_root = RB_ROOT; - ref_tree->unique_refs = 0; - - return ref_tree; -} - -/* Free all nodes in the ref tree, and reinit ref_root */ -static void ref_root_fini(struct ref_root *ref_tree) -{ - struct ref_node *node; - struct rb_node *next; - - while ((next = rb_first(&ref_tree->rb_root)) != NULL) { - node = rb_entry(next, struct ref_node, rb_node); - rb_erase(next, &ref_tree->rb_root); - kfree(node); - } - - ref_tree->rb_root = RB_ROOT; - ref_tree->unique_refs = 0; -} - -static void ref_root_free(struct ref_root *ref_tree) -{ - if (!ref_tree) - return; - - ref_root_fini(ref_tree); - kfree(ref_tree); -} - -/* - * Compare ref_node with (root_id, object_id, offset, parent) - * - * The function compares two ref_node a and b. It returns an integer less - * than, equal to, or greater than zero , respectively, to be less than, to - * equal, or be greater than b. - */ -static int ref_node_cmp(struct ref_node *a, struct ref_node *b) -{ - if (a->root_id < b->root_id) - return -1; - else if (a->root_id > b->root_id) - return 1; - - if (a->object_id < b->object_id) - return -1; - else if (a->object_id > b->object_id) - return 1; - - if (a->offset < b->offset) - return -1; - else if (a->offset > b->offset) - return 1; - - if (a->parent < b->parent) - return -1; - else if (a->parent > b->parent) - return 1; - - return 0; -} - -/* - * Search ref_node with (root_id, object_id, offset, parent) in the tree - * - * if found, the pointer of the ref_node will be returned; - * if not found, NULL will be returned and pos will point to the rb_node for - * insert, pos_parent will point to pos'parent for insert; -*/ -static struct ref_node *__ref_tree_search(struct ref_root *ref_tree, - struct rb_node ***pos, - struct rb_node **pos_parent, - u64 root_id, u64 object_id, - u64 offset, u64 parent) -{ - struct ref_node *cur = NULL; - struct ref_node entry; - int ret; - - entry.root_id = root_id; - entry.object_id = object_id; - entry.offset = offset; - entry.parent = parent; - - *pos = &ref_tree->rb_root.rb_node; - - while (**pos) { - *pos_parent = **pos; - cur = rb_entry(*pos_parent, struct ref_node, rb_node); - - ret = ref_node_cmp(cur, &entry); - if (ret > 0) - *pos = &(**pos)->rb_left; - else if (ret < 0) - *pos = &(**pos)->rb_right; - else - return cur; - } - - return NULL; -} - -/* - * Insert a ref_node to the ref tree - * @pos used for specifiy the position to insert - * @pos_parent for specifiy pos's parent - * - * success, return 0; - * ref_node already exists, return -EEXIST; -*/ -static int ref_tree_insert(struct ref_root *ref_tree, struct rb_node **pos, - struct rb_node *pos_parent, struct ref_node *ins) -{ - struct rb_node **p = NULL; - struct rb_node *parent = NULL; - struct ref_node *cur = NULL; - - if (!pos) { - cur = __ref_tree_search(ref_tree, &p, &parent, ins->root_id, - ins->object_id, ins->offset, - ins->parent); - if (cur) - return -EEXIST; - } else { - p = pos; - parent = pos_parent; - } - - rb_link_node(&ins->rb_node, parent, p); - rb_insert_color(&ins->rb_node, &ref_tree->rb_root); - - return 0; -} - -/* Erase and free ref_node, caller should update ref_root->unique_refs */ -static void ref_tree_remove(struct ref_root *ref_tree, struct ref_node *node) -{ - rb_erase(&node->rb_node, &ref_tree->rb_root); - kfree(node); -} - -/* - * Update ref_root->unique_refs - * - * Call __ref_tree_search - * 1. if ref_node doesn't exist, ref_tree_insert this node, and update - * ref_root->unique_refs: - * if ref_node->ref_mod > 0, ref_root->unique_refs++; - * if ref_node->ref_mod < 0, do noting; - * - * 2. if ref_node is found, then get origin ref_node->ref_mod, and update - * ref_node->ref_mod. - * if ref_node->ref_mod is equal to 0,then call ref_tree_remove - * - * according to origin_mod and new_mod, update ref_root->items - * +----------------+--------------+-------------+ - * | |new_count <= 0|new_count > 0| - * +----------------+--------------+-------------+ - * |origin_count < 0| 0 | 1 | - * +----------------+--------------+-------------+ - * |origin_count > 0| -1 | 0 | - * +----------------+--------------+-------------+ - * - * In case of allocation failure, -ENOMEM is returned and the ref_tree stays - * unaltered. - * Success, return 0 - */ -static int ref_tree_add(struct ref_root *ref_tree, u64 root_id, u64 object_id, - u64 offset, u64 parent, int count) -{ - struct ref_node *node = NULL; - struct rb_node **pos = NULL; - struct rb_node *pos_parent = NULL; - int origin_count; - int ret; - - if (!count) - return 0; - - node = __ref_tree_search(ref_tree, &pos, &pos_parent, root_id, - object_id, offset, parent); - if (node == NULL) { - node = kmalloc(sizeof(*node), GFP_NOFS); - if (!node) - return -ENOMEM; - - node->root_id = root_id; - node->object_id = object_id; - node->offset = offset; - node->parent = parent; - node->ref_mod = count; - - ret = ref_tree_insert(ref_tree, pos, pos_parent, node); - ASSERT(!ret); - if (ret) { - kfree(node); - return ret; - } - - ref_tree->unique_refs += node->ref_mod > 0 ? 1 : 0; - - return 0; - } - - origin_count = node->ref_mod; - node->ref_mod += count; - - if (node->ref_mod > 0) - ref_tree->unique_refs += origin_count > 0 ? 0 : 1; - else if (node->ref_mod <= 0) - ref_tree->unique_refs += origin_count > 0 ? -1 : 0; - - if (!node->ref_mod) - ref_tree_remove(ref_tree, node); - - return 0; -} - -static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb, - struct btrfs_file_extent_item *fi, - u64 extent_item_pos, - struct extent_inode_elem **eie) +static int check_extent_in_eb(struct btrfs_backref_walk_ctx *ctx, + const struct btrfs_key *key, + const struct extent_buffer *eb, + const struct btrfs_file_extent_item *fi, + struct extent_inode_elem **eie) { - u64 offset = 0; + const u64 data_len = btrfs_file_extent_num_bytes(eb, fi); + u64 offset = key->offset; struct extent_inode_elem *e; + const u64 *root_ids; + int root_count; + bool cached; - if (!btrfs_file_extent_compression(eb, fi) && + if (!ctx->ignore_extent_item_pos && + !btrfs_file_extent_compression(eb, fi) && !btrfs_file_extent_encryption(eb, fi) && !btrfs_file_extent_other_encoding(eb, fi)) { u64 data_offset; - u64 data_len; data_offset = btrfs_file_extent_offset(eb, fi); - data_len = btrfs_file_extent_num_bytes(eb, fi); - if (extent_item_pos < data_offset || - extent_item_pos >= data_offset + data_len) + if (ctx->extent_item_pos < data_offset || + ctx->extent_item_pos >= data_offset + data_len) return 1; - offset = extent_item_pos - data_offset; + offset += ctx->extent_item_pos - data_offset; } + if (!ctx->indirect_ref_iterator || !ctx->cache_lookup) + goto add_inode_elem; + + cached = ctx->cache_lookup(eb->start, ctx->user_ctx, &root_ids, + &root_count); + if (!cached) + goto add_inode_elem; + + for (int i = 0; i < root_count; i++) { + int ret; + + ret = ctx->indirect_ref_iterator(key->objectid, offset, + data_len, root_ids[i], + ctx->user_ctx); + if (ret) + return ret; + } + +add_inode_elem: e = kmalloc(sizeof(*e), GFP_NOFS); if (!e) return -ENOMEM; e->next = *eie; e->inum = key->objectid; - e->offset = key->offset + offset; + e->offset = offset; + e->num_bytes = data_len; *eie = e; return 0; @@ -344,9 +101,9 @@ static void free_inode_elem_list(struct extent_inode_elem *eie) } } -static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte, - u64 extent_item_pos, - struct extent_inode_elem **eie) +static int find_extent_in_eb(struct btrfs_backref_walk_ctx *ctx, + const struct extent_buffer *eb, + struct extent_inode_elem **eie) { u64 disk_byte; struct btrfs_key key; @@ -372,50 +129,223 @@ static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte, continue; /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */ disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); - if (disk_byte != wanted_disk_byte) + if (disk_byte != ctx->bytenr) continue; - ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie); - if (ret < 0) + ret = check_extent_in_eb(ctx, &key, eb, fi, eie); + if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0) return ret; } return 0; } +struct preftree { + struct rb_root_cached root; + unsigned int count; +}; + +#define PREFTREE_INIT { .root = RB_ROOT_CACHED, .count = 0 } + +struct preftrees { + struct preftree direct; /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */ + struct preftree indirect; /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */ + struct preftree indirect_missing_keys; +}; + /* - * this structure records all encountered refs on the way up to the root + * Checks for a shared extent during backref search. + * + * The share_count tracks prelim_refs (direct and indirect) having a + * ref->count >0: + * - incremented when a ref->count transitions to >0 + * - decremented when a ref->count transitions to <1 */ -struct __prelim_ref { - struct list_head list; - u64 root_id; - struct btrfs_key key_for_search; - int level; - int count; - struct extent_inode_elem *inode_list; - u64 parent; - u64 wanted_disk_byte; +struct share_check { + struct btrfs_backref_share_check_ctx *ctx; + struct btrfs_root *root; + u64 inum; + u64 data_bytenr; + u64 data_extent_gen; + /* + * Counts number of inodes that refer to an extent (different inodes in + * the same root or different roots) that we could find. The sharedness + * check typically stops once this counter gets greater than 1, so it + * may not reflect the total number of inodes. + */ + int share_count; + /* + * The number of times we found our inode refers to the data extent we + * are determining the sharedness. In other words, how many file extent + * items we could find for our inode that point to our target data + * extent. The value we get here after finishing the extent sharedness + * check may be smaller than reality, but if it ends up being greater + * than 1, then we know for sure the inode has multiple file extent + * items that point to our inode, and we can safely assume it's useful + * to cache the sharedness check result. + */ + int self_ref_count; + bool have_delayed_delete_refs; }; +static inline int extent_is_shared(struct share_check *sc) +{ + return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0; +} + static struct kmem_cache *btrfs_prelim_ref_cache; int __init btrfs_prelim_ref_init(void) { btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref", - sizeof(struct __prelim_ref), - 0, - SLAB_MEM_SPREAD, - NULL); + sizeof(struct prelim_ref), 0, 0, NULL); if (!btrfs_prelim_ref_cache) return -ENOMEM; return 0; } -void btrfs_prelim_ref_exit(void) +void __cold btrfs_prelim_ref_exit(void) { kmem_cache_destroy(btrfs_prelim_ref_cache); } +static void free_pref(struct prelim_ref *ref) +{ + kmem_cache_free(btrfs_prelim_ref_cache, ref); +} + +/* + * Return 0 when both refs are for the same block (and can be merged). + * A -1 return indicates ref1 is a 'lower' block than ref2, while 1 + * indicates a 'higher' block. + */ +static int prelim_ref_compare(const struct prelim_ref *ref1, + const struct prelim_ref *ref2) +{ + if (ref1->level < ref2->level) + return -1; + if (ref1->level > ref2->level) + return 1; + if (ref1->root_id < ref2->root_id) + return -1; + if (ref1->root_id > ref2->root_id) + return 1; + if (ref1->key_for_search.type < ref2->key_for_search.type) + return -1; + if (ref1->key_for_search.type > ref2->key_for_search.type) + return 1; + if (ref1->key_for_search.objectid < ref2->key_for_search.objectid) + return -1; + if (ref1->key_for_search.objectid > ref2->key_for_search.objectid) + return 1; + if (ref1->key_for_search.offset < ref2->key_for_search.offset) + return -1; + if (ref1->key_for_search.offset > ref2->key_for_search.offset) + return 1; + if (ref1->parent < ref2->parent) + return -1; + if (ref1->parent > ref2->parent) + return 1; + + return 0; +} + +static int prelim_ref_rb_add_cmp(const struct rb_node *new, + const struct rb_node *exist) +{ + const struct prelim_ref *ref_new = + rb_entry(new, struct prelim_ref, rbnode); + const struct prelim_ref *ref_exist = + rb_entry(exist, struct prelim_ref, rbnode); + + /* + * prelim_ref_compare() expects the first parameter as the existing one, + * different from the rb_find_add_cached() order. + */ + return prelim_ref_compare(ref_exist, ref_new); +} + +static void update_share_count(struct share_check *sc, int oldcount, + int newcount, const struct prelim_ref *newref) +{ + if ((!sc) || (oldcount == 0 && newcount < 1)) + return; + + if (oldcount > 0 && newcount < 1) + sc->share_count--; + else if (oldcount < 1 && newcount > 0) + sc->share_count++; + + if (newref->root_id == btrfs_root_id(sc->root) && + newref->wanted_disk_byte == sc->data_bytenr && + newref->key_for_search.objectid == sc->inum) + sc->self_ref_count += newref->count; +} + +/* + * Add @newref to the @root rbtree, merging identical refs. + * + * Callers should assume that newref has been freed after calling. + */ +static void prelim_ref_insert(const struct btrfs_fs_info *fs_info, + struct preftree *preftree, + struct prelim_ref *newref, + struct share_check *sc) +{ + struct rb_root_cached *root; + struct rb_node *exist; + + root = &preftree->root; + exist = rb_find_add_cached(&newref->rbnode, root, prelim_ref_rb_add_cmp); + if (exist) { + struct prelim_ref *ref = rb_entry(exist, struct prelim_ref, rbnode); + /* Identical refs, merge them and free @newref */ + struct extent_inode_elem *eie = ref->inode_list; + + while (eie && eie->next) + eie = eie->next; + + if (!eie) + ref->inode_list = newref->inode_list; + else + eie->next = newref->inode_list; + trace_btrfs_prelim_ref_merge(fs_info, ref, newref, + preftree->count); + /* + * A delayed ref can have newref->count < 0. + * The ref->count is updated to follow any + * BTRFS_[ADD|DROP]_DELAYED_REF actions. + */ + update_share_count(sc, ref->count, + ref->count + newref->count, newref); + ref->count += newref->count; + free_pref(newref); + return; + } + + update_share_count(sc, 0, newref->count, newref); + preftree->count++; + trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count); +} + +/* + * Release the entire tree. We don't care about internal consistency so + * just free everything and then reset the tree root. + */ +static void prelim_release(struct preftree *preftree) +{ + struct prelim_ref *ref, *next_ref; + + rbtree_postorder_for_each_entry_safe(ref, next_ref, + &preftree->root.rb_root, rbnode) { + free_inode_elem_list(ref->inode_list); + free_pref(ref); + } + + preftree->root = RB_ROOT_CACHED; + preftree->count = 0; +} + /* * the rules for all callers of this function are: * - obtaining the parent is the goal @@ -448,19 +378,19 @@ void btrfs_prelim_ref_exit(void) * * - column 1, 3: we've the parent -> done * - column 2: we take the first key from the block to find the parent - * (see __add_missing_keys) + * (see add_missing_keys) * - column 4: we use the key to find the parent * * additional information that's available but not required to find the parent * block might help in merging entries to gain some speed. */ - -static int __add_prelim_ref(struct list_head *head, u64 root_id, - struct btrfs_key *key, int level, - u64 parent, u64 wanted_disk_byte, int count, - gfp_t gfp_mask) +static int add_prelim_ref(const struct btrfs_fs_info *fs_info, + struct preftree *preftree, u64 root_id, + const struct btrfs_key *key, int level, u64 parent, + u64 wanted_disk_byte, int count, + struct share_check *sc, gfp_t gfp_mask) { - struct __prelim_ref *ref; + struct prelim_ref *ref; if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID) return 0; @@ -470,48 +400,75 @@ static int __add_prelim_ref(struct list_head *head, u64 root_id, return -ENOMEM; ref->root_id = root_id; - if (key) { + if (key) ref->key_for_search = *key; - /* - * We can often find data backrefs with an offset that is too - * large (>= LLONG_MAX, maximum allowed file offset) due to - * underflows when subtracting a file's offset with the data - * offset of its corresponding extent data item. This can - * happen for example in the clone ioctl. - * So if we detect such case we set the search key's offset to - * zero to make sure we will find the matching file extent item - * at add_all_parents(), otherwise we will miss it because the - * offset taken form the backref is much larger then the offset - * of the file extent item. This can make us scan a very large - * number of file extent items, but at least it will not make - * us miss any. - * This is an ugly workaround for a behaviour that should have - * never existed, but it does and a fix for the clone ioctl - * would touch a lot of places, cause backwards incompatibility - * and would not fix the problem for extents cloned with older - * kernels. - */ - if (ref->key_for_search.type == BTRFS_EXTENT_DATA_KEY && - ref->key_for_search.offset >= LLONG_MAX) - ref->key_for_search.offset = 0; - } else { + else memset(&ref->key_for_search, 0, sizeof(ref->key_for_search)); - } ref->inode_list = NULL; ref->level = level; ref->count = count; ref->parent = parent; ref->wanted_disk_byte = wanted_disk_byte; - list_add_tail(&ref->list, head); + prelim_ref_insert(fs_info, preftree, ref, sc); + return extent_is_shared(sc); +} + +/* direct refs use root == 0, key == NULL */ +static int add_direct_ref(const struct btrfs_fs_info *fs_info, + struct preftrees *preftrees, int level, u64 parent, + u64 wanted_disk_byte, int count, + struct share_check *sc, gfp_t gfp_mask) +{ + return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level, + parent, wanted_disk_byte, count, sc, gfp_mask); +} + +/* indirect refs use parent == 0 */ +static int add_indirect_ref(const struct btrfs_fs_info *fs_info, + struct preftrees *preftrees, u64 root_id, + const struct btrfs_key *key, int level, + u64 wanted_disk_byte, int count, + struct share_check *sc, gfp_t gfp_mask) +{ + struct preftree *tree = &preftrees->indirect; + if (!key) + tree = &preftrees->indirect_missing_keys; + return add_prelim_ref(fs_info, tree, root_id, key, level, 0, + wanted_disk_byte, count, sc, gfp_mask); +} + +static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr) +{ + struct rb_node **p = &preftrees->direct.root.rb_root.rb_node; + struct rb_node *parent = NULL; + struct prelim_ref *ref = NULL; + struct prelim_ref target = {}; + int result; + + target.parent = bytenr; + + while (*p) { + parent = *p; + ref = rb_entry(parent, struct prelim_ref, rbnode); + result = prelim_ref_compare(ref, &target); + + if (result < 0) + p = &(*p)->rb_left; + else if (result > 0) + p = &(*p)->rb_right; + else + return 1; + } return 0; } -static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path, - struct ulist *parents, struct __prelim_ref *ref, - int level, u64 time_seq, const u64 *extent_item_pos, - u64 total_refs) +static int add_all_parents(struct btrfs_backref_walk_ctx *ctx, + struct btrfs_root *root, struct btrfs_path *path, + struct ulist *parents, + struct preftrees *preftrees, struct prelim_ref *ref, + int level) { int ret = 0; int slot; @@ -523,6 +480,8 @@ static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path, u64 disk_byte; u64 wanted_disk_byte = ref->wanted_disk_byte; u64 count = 0; + u64 data_offset; + u8 type; if (level != 0) { eb = path->nodes[level]; @@ -533,18 +492,26 @@ static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path, } /* - * We normally enter this function with the path already pointing to - * the first item to check. But sometimes, we may enter it with - * slot==nritems. In that case, go to the next leaf before we continue. + * 1. We normally enter this function with the path already pointing to + * the first item to check. But sometimes, we may enter it with + * slot == nritems. + * 2. We are searching for normal backref but bytenr of this leaf + * matches shared data backref + * 3. The leaf owner is not equal to the root we are searching + * + * For these cases, go to the next leaf before we continue. */ - if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { - if (time_seq == SEQ_LAST) + eb = path->nodes[0]; + if (path->slots[0] >= btrfs_header_nritems(eb) || + is_shared_data_backref(preftrees, eb->start) || + ref->root_id != btrfs_header_owner(eb)) { + if (ctx->time_seq == BTRFS_SEQ_LAST) ret = btrfs_next_leaf(root, path); else - ret = btrfs_next_old_leaf(root, path, time_seq); + ret = btrfs_next_old_leaf(root, path, ctx->time_seq); } - while (!ret && count < total_refs) { + while (!ret && count < ref->count) { eb = path->nodes[0]; slot = path->slots[0]; @@ -554,18 +521,38 @@ static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path, key.type != BTRFS_EXTENT_DATA_KEY) break; + /* + * We are searching for normal backref but bytenr of this leaf + * matches shared data backref, OR + * the leaf owner is not equal to the root we are searching for + */ + if (slot == 0 && + (is_shared_data_backref(preftrees, eb->start) || + ref->root_id != btrfs_header_owner(eb))) { + if (ctx->time_seq == BTRFS_SEQ_LAST) + ret = btrfs_next_leaf(root, path); + else + ret = btrfs_next_old_leaf(root, path, ctx->time_seq); + continue; + } fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item); + type = btrfs_file_extent_type(eb, fi); + if (type == BTRFS_FILE_EXTENT_INLINE) + goto next; disk_byte = btrfs_file_extent_disk_bytenr(eb, fi); + data_offset = btrfs_file_extent_offset(eb, fi); if (disk_byte == wanted_disk_byte) { eie = NULL; old = NULL; - count++; - if (extent_item_pos) { - ret = check_extent_in_eb(&key, eb, fi, - *extent_item_pos, - &eie); - if (ret < 0) + if (ref->key_for_search.offset == key.offset - data_offset) + count++; + else + goto next; + if (!ctx->skip_inode_ref_list) { + ret = check_extent_in_eb(ctx, &key, eb, fi, &eie); + if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || + ret < 0) break; } if (ret > 0) @@ -574,7 +561,7 @@ static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path, eie, (void **)&old, GFP_NOFS); if (ret < 0) break; - if (!ret && extent_item_pos) { + if (!ret && !ctx->skip_inode_ref_list) { while (old->next) old = old->next; old->next = eie; @@ -582,16 +569,17 @@ static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path, eie = NULL; } next: - if (time_seq == SEQ_LAST) + if (ctx->time_seq == BTRFS_SEQ_LAST) ret = btrfs_next_item(root, path); else - ret = btrfs_next_old_item(root, path, time_seq); + ret = btrfs_next_old_item(root, path, ctx->time_seq); } - if (ret > 0) - ret = 0; - else if (ret < 0) + if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0) free_inode_elem_list(eie); + else if (ret > 0) + ret = 0; + return ret; } @@ -599,67 +587,88 @@ next: * resolve an indirect backref in the form (root_id, key, level) * to a logical address */ -static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info, - struct btrfs_path *path, u64 time_seq, - struct __prelim_ref *ref, - struct ulist *parents, - const u64 *extent_item_pos, u64 total_refs) +static int resolve_indirect_ref(struct btrfs_backref_walk_ctx *ctx, + struct btrfs_path *path, + struct preftrees *preftrees, + struct prelim_ref *ref, struct ulist *parents) { struct btrfs_root *root; - struct btrfs_key root_key; struct extent_buffer *eb; int ret = 0; int root_level; int level = ref->level; - int index; - - root_key.objectid = ref->root_id; - root_key.type = BTRFS_ROOT_ITEM_KEY; - root_key.offset = (u64)-1; + struct btrfs_key search_key = ref->key_for_search; - index = srcu_read_lock(&fs_info->subvol_srcu); - - root = btrfs_get_fs_root(fs_info, &root_key, false); + /* + * If we're search_commit_root we could possibly be holding locks on + * other tree nodes. This happens when qgroups does backref walks when + * adding new delayed refs. To deal with this we need to look in cache + * for the root, and if we don't find it then we need to search the + * tree_root's commit root, thus the btrfs_get_fs_root_commit_root usage + * here. + */ + if (path->search_commit_root) + root = btrfs_get_fs_root_commit_root(ctx->fs_info, path, ref->root_id); + else + root = btrfs_get_fs_root(ctx->fs_info, ref->root_id, false); if (IS_ERR(root)) { - srcu_read_unlock(&fs_info->subvol_srcu, index); ret = PTR_ERR(root); + goto out_free; + } + + if (!path->search_commit_root && + test_bit(BTRFS_ROOT_DELETING, &root->state)) { + ret = -ENOENT; goto out; } - if (btrfs_is_testing(fs_info)) { - srcu_read_unlock(&fs_info->subvol_srcu, index); + if (btrfs_is_testing(ctx->fs_info)) { ret = -ENOENT; goto out; } if (path->search_commit_root) root_level = btrfs_header_level(root->commit_root); - else if (time_seq == SEQ_LAST) + else if (ctx->time_seq == BTRFS_SEQ_LAST) root_level = btrfs_header_level(root->node); else - root_level = btrfs_old_root_level(root, time_seq); + root_level = btrfs_old_root_level(root, ctx->time_seq); - if (root_level + 1 == level) { - srcu_read_unlock(&fs_info->subvol_srcu, index); + if (root_level + 1 == level) goto out; - } + /* + * We can often find data backrefs with an offset that is too large + * (>= LLONG_MAX, maximum allowed file offset) due to underflows when + * subtracting a file's offset with the data offset of its + * corresponding extent data item. This can happen for example in the + * clone ioctl. + * + * So if we detect such case we set the search key's offset to zero to + * make sure we will find the matching file extent item at + * add_all_parents(), otherwise we will miss it because the offset + * taken form the backref is much larger then the offset of the file + * extent item. This can make us scan a very large number of file + * extent items, but at least it will not make us miss any. + * + * This is an ugly workaround for a behaviour that should have never + * existed, but it does and a fix for the clone ioctl would touch a lot + * of places, cause backwards incompatibility and would not fix the + * problem for extents cloned with older kernels. + */ + if (search_key.type == BTRFS_EXTENT_DATA_KEY && + search_key.offset >= LLONG_MAX) + search_key.offset = 0; path->lowest_level = level; - if (time_seq == SEQ_LAST) - ret = btrfs_search_slot(NULL, root, &ref->key_for_search, path, - 0, 0); + if (ctx->time_seq == BTRFS_SEQ_LAST) + ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0); else - ret = btrfs_search_old_slot(root, &ref->key_for_search, path, - time_seq); + ret = btrfs_search_old_slot(root, &search_key, path, ctx->time_seq); - /* root node has been locked, we can release @subvol_srcu safely here */ - srcu_read_unlock(&fs_info->subvol_srcu, index); - - btrfs_debug(fs_info, - "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)", - ref->root_id, level, ref->count, ret, - ref->key_for_search.objectid, ref->key_for_search.type, - ref->key_for_search.offset); + btrfs_debug(ctx->fs_info, +"search slot in root %llu (level %d, ref count %d) returned %d for key " BTRFS_KEY_FMT, + ref->root_id, level, ref->count, ret, + BTRFS_KEY_FMT_VALUE(&ref->key_for_search)); if (ret < 0) goto out; @@ -673,61 +682,108 @@ static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info, eb = path->nodes[level]; } - ret = add_all_parents(root, path, parents, ref, level, time_seq, - extent_item_pos, total_refs); + ret = add_all_parents(ctx, root, path, parents, preftrees, ref, level); out: + btrfs_put_root(root); +out_free: path->lowest_level = 0; btrfs_release_path(path); return ret; } +static struct extent_inode_elem * +unode_aux_to_inode_list(struct ulist_node *node) +{ + if (!node) + return NULL; + return (struct extent_inode_elem *)(uintptr_t)node->aux; +} + +static void free_leaf_list(struct ulist *ulist) +{ + struct ulist_node *node; + struct ulist_iterator uiter; + + ULIST_ITER_INIT(&uiter); + while ((node = ulist_next(ulist, &uiter))) + free_inode_elem_list(unode_aux_to_inode_list(node)); + + ulist_free(ulist); +} + /* - * resolve all indirect backrefs from the list + * We maintain three separate rbtrees: one for direct refs, one for + * indirect refs which have a key, and one for indirect refs which do not + * have a key. Each tree does merge on insertion. + * + * Once all of the references are located, we iterate over the tree of + * indirect refs with missing keys. An appropriate key is located and + * the ref is moved onto the tree for indirect refs. After all missing + * keys are thus located, we iterate over the indirect ref tree, resolve + * each reference, and then insert the resolved reference onto the + * direct tree (merging there too). + * + * New backrefs (i.e., for parent nodes) are added to the appropriate + * rbtree as they are encountered. The new backrefs are subsequently + * resolved as above. */ -static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info, - struct btrfs_path *path, u64 time_seq, - struct list_head *head, - const u64 *extent_item_pos, u64 total_refs, - u64 root_objectid) +static int resolve_indirect_refs(struct btrfs_backref_walk_ctx *ctx, + struct btrfs_path *path, + struct preftrees *preftrees, + struct share_check *sc) { - int err; int ret = 0; - struct __prelim_ref *ref; - struct __prelim_ref *ref_safe; - struct __prelim_ref *new_ref; struct ulist *parents; struct ulist_node *node; struct ulist_iterator uiter; + struct rb_node *rnode; parents = ulist_alloc(GFP_NOFS); if (!parents) return -ENOMEM; /* - * _safe allows us to insert directly after the current item without - * iterating over the newly inserted items. - * we're also allowed to re-assign ref during iteration. + * We could trade memory usage for performance here by iterating + * the tree, allocating new refs for each insertion, and then + * freeing the entire indirect tree when we're done. In some test + * cases, the tree can grow quite large (~200k objects). */ - list_for_each_entry_safe(ref, ref_safe, head, list) { - if (ref->parent) /* already direct */ - continue; - if (ref->count == 0) + while ((rnode = rb_first_cached(&preftrees->indirect.root))) { + struct prelim_ref *ref; + int ret2; + + ref = rb_entry(rnode, struct prelim_ref, rbnode); + if (WARN(ref->parent, + "BUG: direct ref found in indirect tree")) { + ret = -EINVAL; + goto out; + } + + rb_erase_cached(&ref->rbnode, &preftrees->indirect.root); + preftrees->indirect.count--; + + if (ref->count == 0) { + free_pref(ref); continue; - if (root_objectid && ref->root_id != root_objectid) { + } + + if (sc && ref->root_id != btrfs_root_id(sc->root)) { + free_pref(ref); ret = BACKREF_FOUND_SHARED; goto out; } - err = __resolve_indirect_ref(fs_info, path, time_seq, ref, - parents, extent_item_pos, - total_refs); + ret2 = resolve_indirect_ref(ctx, path, preftrees, ref, parents); /* * we can only tolerate ENOENT,otherwise,we should catch error * and return directly. */ - if (err == -ENOENT) { + if (ret2 == -ENOENT) { + prelim_ref_insert(ctx->fs_info, &preftrees->direct, ref, + NULL); continue; - } else if (err) { - ret = err; + } else if (ret2) { + free_pref(ref); + ret = ret2; goto out; } @@ -735,148 +791,112 @@ static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info, ULIST_ITER_INIT(&uiter); node = ulist_next(parents, &uiter); ref->parent = node ? node->val : 0; - ref->inode_list = node ? - (struct extent_inode_elem *)(uintptr_t)node->aux : NULL; + ref->inode_list = unode_aux_to_inode_list(node); - /* additional parents require new refs being added here */ + /* Add a prelim_ref(s) for any other parent(s). */ while ((node = ulist_next(parents, &uiter))) { + struct prelim_ref *new_ref; + new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache, GFP_NOFS); if (!new_ref) { + free_pref(ref); ret = -ENOMEM; goto out; } memcpy(new_ref, ref, sizeof(*ref)); new_ref->parent = node->val; - new_ref->inode_list = (struct extent_inode_elem *) - (uintptr_t)node->aux; - list_add(&new_ref->list, &ref->list); + new_ref->inode_list = unode_aux_to_inode_list(node); + prelim_ref_insert(ctx->fs_info, &preftrees->direct, + new_ref, NULL); } + + /* + * Now it's a direct ref, put it in the direct tree. We must + * do this last because the ref could be merged/freed here. + */ + prelim_ref_insert(ctx->fs_info, &preftrees->direct, ref, NULL); + ulist_reinit(parents); + cond_resched(); } out: - ulist_free(parents); + /* + * We may have inode lists attached to refs in the parents ulist, so we + * must free them before freeing the ulist and its refs. + */ + free_leaf_list(parents); return ret; } -static inline int ref_for_same_block(struct __prelim_ref *ref1, - struct __prelim_ref *ref2) -{ - if (ref1->level != ref2->level) - return 0; - if (ref1->root_id != ref2->root_id) - return 0; - if (ref1->key_for_search.type != ref2->key_for_search.type) - return 0; - if (ref1->key_for_search.objectid != ref2->key_for_search.objectid) - return 0; - if (ref1->key_for_search.offset != ref2->key_for_search.offset) - return 0; - if (ref1->parent != ref2->parent) - return 0; - - return 1; -} - /* * read tree blocks and add keys where required. */ -static int __add_missing_keys(struct btrfs_fs_info *fs_info, - struct list_head *head) +static int add_missing_keys(struct btrfs_fs_info *fs_info, + struct preftrees *preftrees, bool lock) { - struct __prelim_ref *ref; + struct prelim_ref *ref; struct extent_buffer *eb; + struct preftree *tree = &preftrees->indirect_missing_keys; + struct rb_node *node; - list_for_each_entry(ref, head, list) { - if (ref->parent) - continue; - if (ref->key_for_search.type) - continue; + while ((node = rb_first_cached(&tree->root))) { + struct btrfs_tree_parent_check check = { 0 }; + + ref = rb_entry(node, struct prelim_ref, rbnode); + rb_erase_cached(node, &tree->root); + + BUG_ON(ref->parent); /* should not be a direct ref */ + BUG_ON(ref->key_for_search.type); BUG_ON(!ref->wanted_disk_byte); - eb = read_tree_block(fs_info, ref->wanted_disk_byte, 0); + + check.level = ref->level - 1; + check.owner_root = ref->root_id; + + eb = read_tree_block(fs_info, ref->wanted_disk_byte, &check); if (IS_ERR(eb)) { + free_pref(ref); return PTR_ERR(eb); - } else if (!extent_buffer_uptodate(eb)) { + } + if (unlikely(!extent_buffer_uptodate(eb))) { + free_pref(ref); free_extent_buffer(eb); return -EIO; } - btrfs_tree_read_lock(eb); + + if (lock) + btrfs_tree_read_lock(eb); if (btrfs_header_level(eb) == 0) btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0); else btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0); - btrfs_tree_read_unlock(eb); + if (lock) + btrfs_tree_read_unlock(eb); free_extent_buffer(eb); + prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL); + cond_resched(); } return 0; } /* - * merge backrefs and adjust counts accordingly - * - * FIXME: For MERGE_IDENTICAL_KEYS, if we add more keys in __add_prelim_ref - * then we can merge more here. Additionally, we could even add a key - * range for the blocks we looked into to merge even more (-> replace - * unresolved refs by those having a parent). - */ -static void __merge_refs(struct list_head *head, enum merge_mode mode) -{ - struct __prelim_ref *pos1; - - list_for_each_entry(pos1, head, list) { - struct __prelim_ref *pos2 = pos1, *tmp; - - list_for_each_entry_safe_continue(pos2, tmp, head, list) { - struct __prelim_ref *ref1 = pos1, *ref2 = pos2; - struct extent_inode_elem *eie; - - if (!ref_for_same_block(ref1, ref2)) - continue; - if (mode == MERGE_IDENTICAL_KEYS) { - if (!ref1->parent && ref2->parent) - swap(ref1, ref2); - } else { - if (ref1->parent != ref2->parent) - continue; - } - - eie = ref1->inode_list; - while (eie && eie->next) - eie = eie->next; - if (eie) - eie->next = ref2->inode_list; - else - ref1->inode_list = ref2->inode_list; - ref1->count += ref2->count; - - list_del(&ref2->list); - kmem_cache_free(btrfs_prelim_ref_cache, ref2); - cond_resched(); - } - - } -} - -/* * add all currently queued delayed refs from this head whose seq nr is * smaller or equal that seq to the list */ -static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq, - struct list_head *prefs, u64 *total_refs, - u64 inum) +static int add_delayed_refs(const struct btrfs_fs_info *fs_info, + struct btrfs_delayed_ref_head *head, u64 seq, + struct preftrees *preftrees, struct share_check *sc) { struct btrfs_delayed_ref_node *node; - struct btrfs_delayed_extent_op *extent_op = head->extent_op; struct btrfs_key key; - struct btrfs_key op_key = {0}; - int sgn; + struct rb_node *n; + int count; int ret = 0; - if (extent_op && extent_op->update_key) - btrfs_disk_key_to_cpu(&op_key, &extent_op->key); - spin_lock(&head->lock); - list_for_each_entry(node, &head->ref_list, list) { + for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) { + node = rb_entry(n, struct btrfs_delayed_ref_node, + ref_node); if (node->seq > seq) continue; @@ -886,83 +906,106 @@ static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq, WARN_ON(1); continue; case BTRFS_ADD_DELAYED_REF: - sgn = 1; + count = node->ref_mod; break; case BTRFS_DROP_DELAYED_REF: - sgn = -1; + count = node->ref_mod * -1; break; default: - BUG_ON(1); + BUG(); } - *total_refs += (node->ref_mod * sgn); switch (node->type) { case BTRFS_TREE_BLOCK_REF_KEY: { - struct btrfs_delayed_tree_ref *ref; + /* NORMAL INDIRECT METADATA backref */ + struct btrfs_key *key_ptr = NULL; + /* The owner of a tree block ref is the level. */ + int level = btrfs_delayed_ref_owner(node); + + if (head->extent_op && head->extent_op->update_key) { + btrfs_disk_key_to_cpu(&key, &head->extent_op->key); + key_ptr = &key; + } - ref = btrfs_delayed_node_to_tree_ref(node); - ret = __add_prelim_ref(prefs, ref->root, &op_key, - ref->level + 1, 0, node->bytenr, - node->ref_mod * sgn, GFP_ATOMIC); + ret = add_indirect_ref(fs_info, preftrees, node->ref_root, + key_ptr, level + 1, node->bytenr, + count, sc, GFP_ATOMIC); break; } case BTRFS_SHARED_BLOCK_REF_KEY: { - struct btrfs_delayed_tree_ref *ref; + /* + * SHARED DIRECT METADATA backref + * + * The owner of a tree block ref is the level. + */ + int level = btrfs_delayed_ref_owner(node); - ref = btrfs_delayed_node_to_tree_ref(node); - ret = __add_prelim_ref(prefs, 0, NULL, - ref->level + 1, ref->parent, - node->bytenr, - node->ref_mod * sgn, GFP_ATOMIC); + ret = add_direct_ref(fs_info, preftrees, level + 1, + node->parent, node->bytenr, count, + sc, GFP_ATOMIC); break; } case BTRFS_EXTENT_DATA_REF_KEY: { - struct btrfs_delayed_data_ref *ref; - ref = btrfs_delayed_node_to_data_ref(node); - - key.objectid = ref->objectid; + /* NORMAL INDIRECT DATA backref */ + key.objectid = btrfs_delayed_ref_owner(node); key.type = BTRFS_EXTENT_DATA_KEY; - key.offset = ref->offset; + key.offset = btrfs_delayed_ref_offset(node); /* - * Found a inum that doesn't match our known inum, we - * know it's shared. + * If we have a share check context and a reference for + * another inode, we can't exit immediately. This is + * because even if this is a BTRFS_ADD_DELAYED_REF + * reference we may find next a BTRFS_DROP_DELAYED_REF + * which cancels out this ADD reference. + * + * If this is a DROP reference and there was no previous + * ADD reference, then we need to signal that when we + * process references from the extent tree (through + * add_inline_refs() and add_keyed_refs()), we should + * not exit early if we find a reference for another + * inode, because one of the delayed DROP references + * may cancel that reference in the extent tree. */ - if (inum && ref->objectid != inum) { - ret = BACKREF_FOUND_SHARED; - break; - } + if (sc && count < 0) + sc->have_delayed_delete_refs = true; - ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0, - node->bytenr, - node->ref_mod * sgn, GFP_ATOMIC); + ret = add_indirect_ref(fs_info, preftrees, node->ref_root, + &key, 0, node->bytenr, count, sc, + GFP_ATOMIC); break; } case BTRFS_SHARED_DATA_REF_KEY: { - struct btrfs_delayed_data_ref *ref; - - ref = btrfs_delayed_node_to_data_ref(node); - ret = __add_prelim_ref(prefs, 0, NULL, 0, - ref->parent, node->bytenr, - node->ref_mod * sgn, GFP_ATOMIC); + /* SHARED DIRECT FULL backref */ + ret = add_direct_ref(fs_info, preftrees, 0, node->parent, + node->bytenr, count, sc, + GFP_ATOMIC); break; } default: WARN_ON(1); } - if (ret) + /* + * We must ignore BACKREF_FOUND_SHARED until all delayed + * refs have been checked. + */ + if (ret && (ret != BACKREF_FOUND_SHARED)) break; } + if (!ret) + ret = extent_is_shared(sc); + spin_unlock(&head->lock); return ret; } /* * add all inline backrefs for bytenr to the list + * + * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED. */ -static int __add_inline_refs(struct btrfs_path *path, u64 bytenr, - int *info_level, struct list_head *prefs, - struct ref_root *ref_tree, - u64 *total_refs, u64 inum) +static int add_inline_refs(struct btrfs_backref_walk_ctx *ctx, + struct btrfs_path *path, + int *info_level, struct preftrees *preftrees, + struct share_check *sc) { int ret = 0; int slot; @@ -981,12 +1024,16 @@ static int __add_inline_refs(struct btrfs_path *path, u64 bytenr, leaf = path->nodes[0]; slot = path->slots[0]; - item_size = btrfs_item_size_nr(leaf, slot); - BUG_ON(item_size < sizeof(*ei)); - + item_size = btrfs_item_size(leaf, slot); ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); + + if (ctx->check_extent_item) { + ret = ctx->check_extent_item(ctx->bytenr, ei, leaf, ctx->user_ctx); + if (ret) + return ret; + } + flags = btrfs_extent_flags(leaf, ei); - *total_refs += btrfs_extent_refs(leaf, ei); btrfs_item_key_to_cpu(leaf, &found_key, slot); ptr = (unsigned long)(ei + 1); @@ -1012,14 +1059,18 @@ static int __add_inline_refs(struct btrfs_path *path, u64 bytenr, int type; iref = (struct btrfs_extent_inline_ref *)ptr; - type = btrfs_extent_inline_ref_type(leaf, iref); + type = btrfs_get_extent_inline_ref_type(leaf, iref, + BTRFS_REF_TYPE_ANY); + if (unlikely(type == BTRFS_REF_TYPE_INVALID)) + return -EUCLEAN; + offset = btrfs_extent_inline_ref_offset(leaf, iref); switch (type) { case BTRFS_SHARED_BLOCK_REF_KEY: - ret = __add_prelim_ref(prefs, 0, NULL, - *info_level + 1, offset, - bytenr, 1, GFP_NOFS); + ret = add_direct_ref(ctx->fs_info, preftrees, + *info_level + 1, offset, + ctx->bytenr, 1, NULL, GFP_NOFS); break; case BTRFS_SHARED_DATA_REF_KEY: { struct btrfs_shared_data_ref *sdref; @@ -1027,21 +1078,15 @@ static int __add_inline_refs(struct btrfs_path *path, u64 bytenr, sdref = (struct btrfs_shared_data_ref *)(iref + 1); count = btrfs_shared_data_ref_count(leaf, sdref); - ret = __add_prelim_ref(prefs, 0, NULL, 0, offset, - bytenr, count, GFP_NOFS); - if (ref_tree) { - if (!ret) - ret = ref_tree_add(ref_tree, 0, 0, 0, - bytenr, count); - if (!ret && ref_tree->unique_refs > 1) - ret = BACKREF_FOUND_SHARED; - } + + ret = add_direct_ref(ctx->fs_info, preftrees, 0, offset, + ctx->bytenr, count, sc, GFP_NOFS); break; } case BTRFS_TREE_BLOCK_REF_KEY: - ret = __add_prelim_ref(prefs, offset, NULL, - *info_level + 1, 0, - bytenr, 1, GFP_NOFS); + ret = add_indirect_ref(ctx->fs_info, preftrees, offset, + NULL, *info_level + 1, + ctx->bytenr, 1, NULL, GFP_NOFS); break; case BTRFS_EXTENT_DATA_REF_KEY: { struct btrfs_extent_data_ref *dref; @@ -1055,25 +1100,25 @@ static int __add_inline_refs(struct btrfs_path *path, u64 bytenr, key.type = BTRFS_EXTENT_DATA_KEY; key.offset = btrfs_extent_data_ref_offset(leaf, dref); - if (inum && key.objectid != inum) { + if (sc && key.objectid != sc->inum && + !sc->have_delayed_delete_refs) { ret = BACKREF_FOUND_SHARED; break; } root = btrfs_extent_data_ref_root(leaf, dref); - ret = __add_prelim_ref(prefs, root, &key, 0, 0, - bytenr, count, GFP_NOFS); - if (ref_tree) { - if (!ret) - ret = ref_tree_add(ref_tree, root, - key.objectid, - key.offset, 0, - count); - if (!ret && ref_tree->unique_refs > 1) - ret = BACKREF_FOUND_SHARED; - } + + if (!ctx->skip_data_ref || + !ctx->skip_data_ref(root, key.objectid, key.offset, + ctx->user_ctx)) + ret = add_indirect_ref(ctx->fs_info, preftrees, + root, &key, 0, ctx->bytenr, + count, sc, GFP_NOFS); break; } + case BTRFS_EXTENT_OWNER_REF_KEY: + ASSERT(btrfs_fs_incompat(ctx->fs_info, SIMPLE_QUOTA)); + break; default: WARN_ON(1); } @@ -1087,13 +1132,16 @@ static int __add_inline_refs(struct btrfs_path *path, u64 bytenr, /* * add all non-inline backrefs for bytenr to the list + * + * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED. */ -static int __add_keyed_refs(struct btrfs_fs_info *fs_info, - struct btrfs_path *path, u64 bytenr, - int info_level, struct list_head *prefs, - struct ref_root *ref_tree, u64 inum) +static int add_keyed_refs(struct btrfs_backref_walk_ctx *ctx, + struct btrfs_root *extent_root, + struct btrfs_path *path, + int info_level, struct preftrees *preftrees, + struct share_check *sc) { - struct btrfs_root *extent_root = fs_info->extent_root; + struct btrfs_fs_info *fs_info = extent_root->fs_info; int ret; int slot; struct extent_buffer *leaf; @@ -1112,7 +1160,7 @@ static int __add_keyed_refs(struct btrfs_fs_info *fs_info, leaf = path->nodes[0]; btrfs_item_key_to_cpu(leaf, &key, slot); - if (key.objectid != bytenr) + if (key.objectid != ctx->bytenr) break; if (key.type < BTRFS_TREE_BLOCK_REF_KEY) continue; @@ -1121,34 +1169,32 @@ static int __add_keyed_refs(struct btrfs_fs_info *fs_info, switch (key.type) { case BTRFS_SHARED_BLOCK_REF_KEY: - ret = __add_prelim_ref(prefs, 0, NULL, - info_level + 1, key.offset, - bytenr, 1, GFP_NOFS); + /* SHARED DIRECT METADATA backref */ + ret = add_direct_ref(fs_info, preftrees, + info_level + 1, key.offset, + ctx->bytenr, 1, NULL, GFP_NOFS); break; case BTRFS_SHARED_DATA_REF_KEY: { + /* SHARED DIRECT FULL backref */ struct btrfs_shared_data_ref *sdref; int count; sdref = btrfs_item_ptr(leaf, slot, struct btrfs_shared_data_ref); count = btrfs_shared_data_ref_count(leaf, sdref); - ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset, - bytenr, count, GFP_NOFS); - if (ref_tree) { - if (!ret) - ret = ref_tree_add(ref_tree, 0, 0, 0, - bytenr, count); - if (!ret && ref_tree->unique_refs > 1) - ret = BACKREF_FOUND_SHARED; - } + ret = add_direct_ref(fs_info, preftrees, 0, + key.offset, ctx->bytenr, count, + sc, GFP_NOFS); break; } case BTRFS_TREE_BLOCK_REF_KEY: - ret = __add_prelim_ref(prefs, key.offset, NULL, - info_level + 1, 0, - bytenr, 1, GFP_NOFS); + /* NORMAL INDIRECT METADATA backref */ + ret = add_indirect_ref(fs_info, preftrees, key.offset, + NULL, info_level + 1, ctx->bytenr, + 1, NULL, GFP_NOFS); break; case BTRFS_EXTENT_DATA_REF_KEY: { + /* NORMAL INDIRECT DATA backref */ struct btrfs_extent_data_ref *dref; int count; u64 root; @@ -1161,23 +1207,20 @@ static int __add_keyed_refs(struct btrfs_fs_info *fs_info, key.type = BTRFS_EXTENT_DATA_KEY; key.offset = btrfs_extent_data_ref_offset(leaf, dref); - if (inum && key.objectid != inum) { + if (sc && key.objectid != sc->inum && + !sc->have_delayed_delete_refs) { ret = BACKREF_FOUND_SHARED; break; } root = btrfs_extent_data_ref_root(leaf, dref); - ret = __add_prelim_ref(prefs, root, &key, 0, 0, - bytenr, count, GFP_NOFS); - if (ref_tree) { - if (!ret) - ret = ref_tree_add(ref_tree, root, - key.objectid, - key.offset, 0, - count); - if (!ret && ref_tree->unique_refs > 1) - ret = BACKREF_FOUND_SHARED; - } + + if (!ctx->skip_data_ref || + !ctx->skip_data_ref(root, key.objectid, key.offset, + ctx->user_ctx)) + ret = add_indirect_ref(fs_info, preftrees, root, + &key, 0, ctx->bytenr, + count, sc, GFP_NOFS); break; } default: @@ -1192,104 +1235,216 @@ static int __add_keyed_refs(struct btrfs_fs_info *fs_info, } /* + * The caller has joined a transaction or is holding a read lock on the + * fs_info->commit_root_sem semaphore, so no need to worry about the root's last + * snapshot field changing while updating or checking the cache. + */ +static bool lookup_backref_shared_cache(struct btrfs_backref_share_check_ctx *ctx, + struct btrfs_root *root, + u64 bytenr, int level, bool *is_shared) +{ + const struct btrfs_fs_info *fs_info = root->fs_info; + struct btrfs_backref_shared_cache_entry *entry; + + if (!current->journal_info) + lockdep_assert_held(&fs_info->commit_root_sem); + + if (!ctx->use_path_cache) + return false; + + if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL)) + return false; + + /* + * Level -1 is used for the data extent, which is not reliable to cache + * because its reference count can increase or decrease without us + * realizing. We cache results only for extent buffers that lead from + * the root node down to the leaf with the file extent item. + */ + ASSERT(level >= 0); + + entry = &ctx->path_cache_entries[level]; + + /* Unused cache entry or being used for some other extent buffer. */ + if (entry->bytenr != bytenr) + return false; + + /* + * We cached a false result, but the last snapshot generation of the + * root changed, so we now have a snapshot. Don't trust the result. + */ + if (!entry->is_shared && + entry->gen != btrfs_root_last_snapshot(&root->root_item)) + return false; + + /* + * If we cached a true result and the last generation used for dropping + * a root changed, we can not trust the result, because the dropped root + * could be a snapshot sharing this extent buffer. + */ + if (entry->is_shared && + entry->gen != btrfs_get_last_root_drop_gen(fs_info)) + return false; + + *is_shared = entry->is_shared; + /* + * If the node at this level is shared, than all nodes below are also + * shared. Currently some of the nodes below may be marked as not shared + * because we have just switched from one leaf to another, and switched + * also other nodes above the leaf and below the current level, so mark + * them as shared. + */ + if (*is_shared) { + for (int i = 0; i < level; i++) { + ctx->path_cache_entries[i].is_shared = true; + ctx->path_cache_entries[i].gen = entry->gen; + } + } + + return true; +} + +/* + * The caller has joined a transaction or is holding a read lock on the + * fs_info->commit_root_sem semaphore, so no need to worry about the root's last + * snapshot field changing while updating or checking the cache. + */ +static void store_backref_shared_cache(struct btrfs_backref_share_check_ctx *ctx, + struct btrfs_root *root, + u64 bytenr, int level, bool is_shared) +{ + const struct btrfs_fs_info *fs_info = root->fs_info; + struct btrfs_backref_shared_cache_entry *entry; + u64 gen; + + if (!current->journal_info) + lockdep_assert_held(&fs_info->commit_root_sem); + + if (!ctx->use_path_cache) + return; + + if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL)) + return; + + /* + * Level -1 is used for the data extent, which is not reliable to cache + * because its reference count can increase or decrease without us + * realizing. We cache results only for extent buffers that lead from + * the root node down to the leaf with the file extent item. + */ + ASSERT(level >= 0); + + if (is_shared) + gen = btrfs_get_last_root_drop_gen(fs_info); + else + gen = btrfs_root_last_snapshot(&root->root_item); + + entry = &ctx->path_cache_entries[level]; + entry->bytenr = bytenr; + entry->is_shared = is_shared; + entry->gen = gen; + + /* + * If we found an extent buffer is shared, set the cache result for all + * extent buffers below it to true. As nodes in the path are COWed, + * their sharedness is moved to their children, and if a leaf is COWed, + * then the sharedness of a data extent becomes direct, the refcount of + * data extent is increased in the extent item at the extent tree. + */ + if (is_shared) { + for (int i = 0; i < level; i++) { + entry = &ctx->path_cache_entries[i]; + entry->is_shared = is_shared; + entry->gen = gen; + } + } +} + +/* * this adds all existing backrefs (inline backrefs, backrefs and delayed * refs) for the given bytenr to the refs list, merges duplicates and resolves * indirect refs to their parent bytenr. * When roots are found, they're added to the roots list * - * NOTE: This can return values > 0 - * - * If time_seq is set to SEQ_LAST, it will not search delayed_refs, and behave - * much like trans == NULL case, the difference only lies in it will not - * commit root. - * The special case is for qgroup to search roots in commit_transaction(). + * @ctx: Backref walking context object, must be not NULL. + * @sc: If !NULL, then immediately return BACKREF_FOUND_SHARED when a + * shared extent is detected. * - * If check_shared is set to 1, any extent has more than one ref item, will - * be returned BACKREF_FOUND_SHARED immediately. + * Otherwise this returns 0 for success and <0 for an error. * * FIXME some caching might speed things up */ -static int find_parent_nodes(struct btrfs_trans_handle *trans, - struct btrfs_fs_info *fs_info, u64 bytenr, - u64 time_seq, struct ulist *refs, - struct ulist *roots, const u64 *extent_item_pos, - u64 root_objectid, u64 inum, int check_shared) +static int find_parent_nodes(struct btrfs_backref_walk_ctx *ctx, + struct share_check *sc) { + struct btrfs_root *root = btrfs_extent_root(ctx->fs_info, ctx->bytenr); struct btrfs_key key; struct btrfs_path *path; struct btrfs_delayed_ref_root *delayed_refs = NULL; struct btrfs_delayed_ref_head *head; int info_level = 0; int ret; - struct list_head prefs_delayed; - struct list_head prefs; - struct __prelim_ref *ref; + struct prelim_ref *ref; + struct rb_node *node; struct extent_inode_elem *eie = NULL; - struct ref_root *ref_tree = NULL; - u64 total_refs = 0; - - INIT_LIST_HEAD(&prefs); - INIT_LIST_HEAD(&prefs_delayed); - - key.objectid = bytenr; - key.offset = (u64)-1; - if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) + struct preftrees preftrees = { + .direct = PREFTREE_INIT, + .indirect = PREFTREE_INIT, + .indirect_missing_keys = PREFTREE_INIT + }; + + /* Roots ulist is not needed when using a sharedness check context. */ + if (sc) + ASSERT(ctx->roots == NULL); + + key.objectid = ctx->bytenr; + if (btrfs_fs_incompat(ctx->fs_info, SKINNY_METADATA)) key.type = BTRFS_METADATA_ITEM_KEY; else key.type = BTRFS_EXTENT_ITEM_KEY; + key.offset = (u64)-1; path = btrfs_alloc_path(); if (!path) return -ENOMEM; - if (!trans) { - path->search_commit_root = 1; - path->skip_locking = 1; + if (!ctx->trans) { + path->search_commit_root = true; + path->skip_locking = true; } - if (time_seq == SEQ_LAST) - path->skip_locking = 1; + if (ctx->time_seq == BTRFS_SEQ_LAST) + path->skip_locking = true; - /* - * grab both a lock on the path and a lock on the delayed ref head. - * We need both to get a consistent picture of how the refs look - * at a specified point in time - */ again: head = NULL; - if (check_shared) { - if (!ref_tree) { - ref_tree = ref_root_alloc(); - if (!ref_tree) { - ret = -ENOMEM; - goto out; - } - } else { - ref_root_fini(ref_tree); - } - } - - ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); + ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); if (ret < 0) goto out; - BUG_ON(ret == 0); - -#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS - if (trans && likely(trans->type != __TRANS_DUMMY) && - time_seq != SEQ_LAST) { -#else - if (trans && time_seq != SEQ_LAST) { -#endif + if (unlikely(ret == 0)) { /* - * look if there are updates for this ref queued and lock the - * head + * Key with offset -1 found, there would have to exist an extent + * item with such offset, but this is out of the valid range. */ - delayed_refs = &trans->transaction->delayed_refs; + ret = -EUCLEAN; + goto out; + } + + if (ctx->trans && likely(ctx->trans->type != __TRANS_DUMMY) && + ctx->time_seq != BTRFS_SEQ_LAST) { + /* + * We have a specific time_seq we care about and trans which + * means we have the path lock, we need to grab the ref head and + * lock it so we have a consistent view of the refs at the given + * time. + */ + delayed_refs = &ctx->trans->transaction->delayed_refs; spin_lock(&delayed_refs->lock); - head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); + head = btrfs_find_delayed_ref_head(ctx->fs_info, delayed_refs, + ctx->bytenr); if (head) { if (!mutex_trylock(&head->mutex)) { - refcount_inc(&head->node.refs); + refcount_inc(&head->refs); spin_unlock(&delayed_refs->lock); btrfs_release_path(path); @@ -1300,49 +1455,18 @@ again: */ mutex_lock(&head->mutex); mutex_unlock(&head->mutex); - btrfs_put_delayed_ref(&head->node); + btrfs_put_delayed_ref_head(head); goto again; } spin_unlock(&delayed_refs->lock); - ret = __add_delayed_refs(head, time_seq, - &prefs_delayed, &total_refs, - inum); + ret = add_delayed_refs(ctx->fs_info, head, ctx->time_seq, + &preftrees, sc); mutex_unlock(&head->mutex); if (ret) goto out; } else { spin_unlock(&delayed_refs->lock); } - - if (check_shared && !list_empty(&prefs_delayed)) { - /* - * Add all delay_ref to the ref_tree and check if there - * are multiple ref items added. - */ - list_for_each_entry(ref, &prefs_delayed, list) { - if (ref->key_for_search.type) { - ret = ref_tree_add(ref_tree, - ref->root_id, - ref->key_for_search.objectid, - ref->key_for_search.offset, - 0, ref->count); - if (ret) - goto out; - } else { - ret = ref_tree_add(ref_tree, 0, 0, 0, - ref->parent, ref->count); - if (ret) - goto out; - } - - } - - if (ref_tree->unique_refs > 1) { - ret = BACKREF_FOUND_SHARED; - goto out; - } - - } } if (path->slots[0]) { @@ -1353,159 +1477,238 @@ again: leaf = path->nodes[0]; slot = path->slots[0]; btrfs_item_key_to_cpu(leaf, &key, slot); - if (key.objectid == bytenr && + if (key.objectid == ctx->bytenr && (key.type == BTRFS_EXTENT_ITEM_KEY || key.type == BTRFS_METADATA_ITEM_KEY)) { - ret = __add_inline_refs(path, bytenr, - &info_level, &prefs, - ref_tree, &total_refs, - inum); + ret = add_inline_refs(ctx, path, &info_level, + &preftrees, sc); if (ret) goto out; - ret = __add_keyed_refs(fs_info, path, bytenr, - info_level, &prefs, - ref_tree, inum); + ret = add_keyed_refs(ctx, root, path, info_level, + &preftrees, sc); if (ret) goto out; } } - btrfs_release_path(path); - list_splice_init(&prefs_delayed, &prefs); + /* + * If we have a share context and we reached here, it means the extent + * is not directly shared (no multiple reference items for it), + * otherwise we would have exited earlier with a return value of + * BACKREF_FOUND_SHARED after processing delayed references or while + * processing inline or keyed references from the extent tree. + * The extent may however be indirectly shared through shared subtrees + * as a result from creating snapshots, so we determine below what is + * its parent node, in case we are dealing with a metadata extent, or + * what's the leaf (or leaves), from a fs tree, that has a file extent + * item pointing to it in case we are dealing with a data extent. + */ + ASSERT(extent_is_shared(sc) == 0); + + /* + * If we are here for a data extent and we have a share_check structure + * it means the data extent is not directly shared (does not have + * multiple reference items), so we have to check if a path in the fs + * tree (going from the root node down to the leaf that has the file + * extent item pointing to the data extent) is shared, that is, if any + * of the extent buffers in the path is referenced by other trees. + */ + if (sc && ctx->bytenr == sc->data_bytenr) { + /* + * If our data extent is from a generation more recent than the + * last generation used to snapshot the root, then we know that + * it can not be shared through subtrees, so we can skip + * resolving indirect references, there's no point in + * determining the extent buffers for the path from the fs tree + * root node down to the leaf that has the file extent item that + * points to the data extent. + */ + if (sc->data_extent_gen > + btrfs_root_last_snapshot(&sc->root->root_item)) { + ret = BACKREF_FOUND_NOT_SHARED; + goto out; + } - ret = __add_missing_keys(fs_info, &prefs); + /* + * If we are only determining if a data extent is shared or not + * and the corresponding file extent item is located in the same + * leaf as the previous file extent item, we can skip resolving + * indirect references for a data extent, since the fs tree path + * is the same (same leaf, so same path). We skip as long as the + * cached result for the leaf is valid and only if there's only + * one file extent item pointing to the data extent, because in + * the case of multiple file extent items, they may be located + * in different leaves and therefore we have multiple paths. + */ + if (sc->ctx->curr_leaf_bytenr == sc->ctx->prev_leaf_bytenr && + sc->self_ref_count == 1) { + bool cached; + bool is_shared; + + cached = lookup_backref_shared_cache(sc->ctx, sc->root, + sc->ctx->curr_leaf_bytenr, + 0, &is_shared); + if (cached) { + if (is_shared) + ret = BACKREF_FOUND_SHARED; + else + ret = BACKREF_FOUND_NOT_SHARED; + goto out; + } + } + } + + btrfs_release_path(path); + + ret = add_missing_keys(ctx->fs_info, &preftrees, !path->skip_locking); if (ret) goto out; - __merge_refs(&prefs, MERGE_IDENTICAL_KEYS); + WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root)); - ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs, - extent_item_pos, total_refs, - root_objectid); + ret = resolve_indirect_refs(ctx, path, &preftrees, sc); if (ret) goto out; - __merge_refs(&prefs, MERGE_IDENTICAL_PARENTS); - - while (!list_empty(&prefs)) { - ref = list_first_entry(&prefs, struct __prelim_ref, list); - WARN_ON(ref->count < 0); - if (roots && ref->count && ref->root_id && ref->parent == 0) { - if (root_objectid && ref->root_id != root_objectid) { - ret = BACKREF_FOUND_SHARED; - goto out; - } + WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root)); + /* + * This walks the tree of merged and resolved refs. Tree blocks are + * read in as needed. Unique entries are added to the ulist, and + * the list of found roots is updated. + * + * We release the entire tree in one go before returning. + */ + node = rb_first_cached(&preftrees.direct.root); + while (node) { + ref = rb_entry(node, struct prelim_ref, rbnode); + node = rb_next(&ref->rbnode); + /* + * ref->count < 0 can happen here if there are delayed + * refs with a node->action of BTRFS_DROP_DELAYED_REF. + * prelim_ref_insert() relies on this when merging + * identical refs to keep the overall count correct. + * prelim_ref_insert() will merge only those refs + * which compare identically. Any refs having + * e.g. different offsets would not be merged, + * and would retain their original ref->count < 0. + */ + if (ctx->roots && ref->count && ref->root_id && ref->parent == 0) { /* no parent == root of tree */ - ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS); + ret = ulist_add(ctx->roots, ref->root_id, 0, GFP_NOFS); if (ret < 0) goto out; } if (ref->count && ref->parent) { - if (extent_item_pos && !ref->inode_list && + if (!ctx->skip_inode_ref_list && !ref->inode_list && ref->level == 0) { + struct btrfs_tree_parent_check check = { 0 }; struct extent_buffer *eb; - eb = read_tree_block(fs_info, ref->parent, 0); + check.level = ref->level; + + eb = read_tree_block(ctx->fs_info, ref->parent, + &check); if (IS_ERR(eb)) { ret = PTR_ERR(eb); goto out; - } else if (!extent_buffer_uptodate(eb)) { + } + if (unlikely(!extent_buffer_uptodate(eb))) { free_extent_buffer(eb); ret = -EIO; goto out; } - btrfs_tree_read_lock(eb); - btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); - ret = find_extent_in_eb(eb, bytenr, - *extent_item_pos, &eie); - btrfs_tree_read_unlock_blocking(eb); + + if (!path->skip_locking) + btrfs_tree_read_lock(eb); + ret = find_extent_in_eb(ctx, eb, &eie); + if (!path->skip_locking) + btrfs_tree_read_unlock(eb); free_extent_buffer(eb); - if (ret < 0) + if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || + ret < 0) goto out; ref->inode_list = eie; + /* + * We transferred the list ownership to the ref, + * so set to NULL to avoid a double free in case + * an error happens after this. + */ + eie = NULL; } - ret = ulist_add_merge_ptr(refs, ref->parent, + ret = ulist_add_merge_ptr(ctx->refs, ref->parent, ref->inode_list, (void **)&eie, GFP_NOFS); if (ret < 0) goto out; - if (!ret && extent_item_pos) { + if (!ret && !ctx->skip_inode_ref_list) { /* - * we've recorded that parent, so we must extend - * its inode list here + * We've recorded that parent, so we must extend + * its inode list here. + * + * However if there was corruption we may not + * have found an eie, return an error in this + * case. */ - BUG_ON(!eie); + ASSERT(eie); + if (unlikely(!eie)) { + ret = -EUCLEAN; + goto out; + } while (eie->next) eie = eie->next; eie->next = ref->inode_list; } eie = NULL; + /* + * We have transferred the inode list ownership from + * this ref to the ref we added to the 'refs' ulist. + * So set this ref's inode list to NULL to avoid + * use-after-free when our caller uses it or double + * frees in case an error happens before we return. + */ + ref->inode_list = NULL; } - list_del(&ref->list); - kmem_cache_free(btrfs_prelim_ref_cache, ref); + cond_resched(); } out: btrfs_free_path(path); - ref_root_free(ref_tree); - while (!list_empty(&prefs)) { - ref = list_first_entry(&prefs, struct __prelim_ref, list); - list_del(&ref->list); - kmem_cache_free(btrfs_prelim_ref_cache, ref); - } - while (!list_empty(&prefs_delayed)) { - ref = list_first_entry(&prefs_delayed, struct __prelim_ref, - list); - list_del(&ref->list); - kmem_cache_free(btrfs_prelim_ref_cache, ref); - } - if (ret < 0) - free_inode_elem_list(eie); - return ret; -} -static void free_leaf_list(struct ulist *blocks) -{ - struct ulist_node *node = NULL; - struct extent_inode_elem *eie; - struct ulist_iterator uiter; + prelim_release(&preftrees.direct); + prelim_release(&preftrees.indirect); + prelim_release(&preftrees.indirect_missing_keys); - ULIST_ITER_INIT(&uiter); - while ((node = ulist_next(blocks, &uiter))) { - if (!node->aux) - continue; - eie = (struct extent_inode_elem *)(uintptr_t)node->aux; + if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0) free_inode_elem_list(eie); - node->aux = 0; - } - - ulist_free(blocks); + return ret; } /* - * Finds all leafs with a reference to the specified combination of bytenr and - * offset. key_list_head will point to a list of corresponding keys (caller must - * free each list element). The leafs will be stored in the leafs ulist, which - * must be freed with ulist_free. + * Finds all leaves with a reference to the specified combination of + * @ctx->bytenr and @ctx->extent_item_pos. The bytenr of the found leaves are + * added to the ulist at @ctx->refs, and that ulist is allocated by this + * function. The caller should free the ulist with free_leaf_list() if + * @ctx->ignore_extent_item_pos is false, otherwise a simple ulist_free() is + * enough. * - * returns 0 on success, <0 on error + * Returns 0 on success and < 0 on error. On error @ctx->refs is not allocated. */ -static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans, - struct btrfs_fs_info *fs_info, u64 bytenr, - u64 time_seq, struct ulist **leafs, - const u64 *extent_item_pos) +int btrfs_find_all_leafs(struct btrfs_backref_walk_ctx *ctx) { int ret; - *leafs = ulist_alloc(GFP_NOFS); - if (!*leafs) + ASSERT(ctx->refs == NULL); + + ctx->refs = ulist_alloc(GFP_NOFS); + if (!ctx->refs) return -ENOMEM; - ret = find_parent_nodes(trans, fs_info, bytenr, time_seq, - *leafs, NULL, extent_item_pos, 0, 0, 0); - if (ret < 0 && ret != -ENOENT) { - free_leaf_list(*leafs); + ret = find_parent_nodes(ctx, NULL); + if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || + (ret < 0 && ret != -ENOENT)) { + free_leaf_list(ctx->refs); + ctx->refs = NULL; return ret; } @@ -1513,7 +1716,7 @@ static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans, } /* - * walk all backrefs for a given extent to find all roots that reference this + * Walk all backrefs for a given extent to find all roots that reference this * extent. Walking a backref means finding all extents that reference this * extent and in turn walk the backrefs of those, too. Naturally this is a * recursive process, but here it is implemented in an iterative fashion: We @@ -1521,122 +1724,318 @@ static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans, * list. In turn, we find all referencing extents for those, further appending * to the list. The way we iterate the list allows adding more elements after * the current while iterating. The process stops when we reach the end of the - * list. Found roots are added to the roots list. + * list. * - * returns 0 on success, < 0 on error. + * Found roots are added to @ctx->roots, which is allocated by this function if + * it points to NULL, in which case the caller is responsible for freeing it + * after it's not needed anymore. + * This function requires @ctx->refs to be NULL, as it uses it for allocating a + * ulist to do temporary work, and frees it before returning. + * + * Returns 0 on success, < 0 on error. */ -static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans, - struct btrfs_fs_info *fs_info, u64 bytenr, - u64 time_seq, struct ulist **roots) +static int btrfs_find_all_roots_safe(struct btrfs_backref_walk_ctx *ctx) { - struct ulist *tmp; - struct ulist_node *node = NULL; + const u64 orig_bytenr = ctx->bytenr; + const bool orig_skip_inode_ref_list = ctx->skip_inode_ref_list; + bool roots_ulist_allocated = false; struct ulist_iterator uiter; - int ret; + int ret = 0; - tmp = ulist_alloc(GFP_NOFS); - if (!tmp) - return -ENOMEM; - *roots = ulist_alloc(GFP_NOFS); - if (!*roots) { - ulist_free(tmp); + ASSERT(ctx->refs == NULL); + + ctx->refs = ulist_alloc(GFP_NOFS); + if (!ctx->refs) return -ENOMEM; + + if (!ctx->roots) { + ctx->roots = ulist_alloc(GFP_NOFS); + if (!ctx->roots) { + ulist_free(ctx->refs); + ctx->refs = NULL; + return -ENOMEM; + } + roots_ulist_allocated = true; } + ctx->skip_inode_ref_list = true; + ULIST_ITER_INIT(&uiter); while (1) { - ret = find_parent_nodes(trans, fs_info, bytenr, time_seq, - tmp, *roots, NULL, 0, 0, 0); + struct ulist_node *node; + + ret = find_parent_nodes(ctx, NULL); if (ret < 0 && ret != -ENOENT) { - ulist_free(tmp); - ulist_free(*roots); - return ret; + if (roots_ulist_allocated) { + ulist_free(ctx->roots); + ctx->roots = NULL; + } + break; } - node = ulist_next(tmp, &uiter); + ret = 0; + node = ulist_next(ctx->refs, &uiter); if (!node) break; - bytenr = node->val; + ctx->bytenr = node->val; cond_resched(); } - ulist_free(tmp); - return 0; + ulist_free(ctx->refs); + ctx->refs = NULL; + ctx->bytenr = orig_bytenr; + ctx->skip_inode_ref_list = orig_skip_inode_ref_list; + + return ret; } -int btrfs_find_all_roots(struct btrfs_trans_handle *trans, - struct btrfs_fs_info *fs_info, u64 bytenr, - u64 time_seq, struct ulist **roots) +int btrfs_find_all_roots(struct btrfs_backref_walk_ctx *ctx, + bool skip_commit_root_sem) { int ret; - if (!trans) - down_read(&fs_info->commit_root_sem); - ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots); - if (!trans) - up_read(&fs_info->commit_root_sem); + if (!ctx->trans && !skip_commit_root_sem) + down_read(&ctx->fs_info->commit_root_sem); + ret = btrfs_find_all_roots_safe(ctx); + if (!ctx->trans && !skip_commit_root_sem) + up_read(&ctx->fs_info->commit_root_sem); return ret; } -/** - * btrfs_check_shared - tell us whether an extent is shared +struct btrfs_backref_share_check_ctx *btrfs_alloc_backref_share_check_ctx(void) +{ + struct btrfs_backref_share_check_ctx *ctx; + + ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); + if (!ctx) + return NULL; + + ulist_init(&ctx->refs); + + return ctx; +} + +void btrfs_free_backref_share_ctx(struct btrfs_backref_share_check_ctx *ctx) +{ + if (!ctx) + return; + + ulist_release(&ctx->refs); + kfree(ctx); +} + +/* + * Check if a data extent is shared or not. * - * @trans: optional trans handle + * @inode: The inode whose extent we are checking. + * @bytenr: Logical bytenr of the extent we are checking. + * @extent_gen: Generation of the extent (file extent item) or 0 if it is + * not known. + * @ctx: A backref sharedness check context. * - * btrfs_check_shared uses the backref walking code but will short + * btrfs_is_data_extent_shared uses the backref walking code but will short * circuit as soon as it finds a root or inode that doesn't match the * one passed in. This provides a significant performance benefit for * callers (such as fiemap) which want to know whether the extent is * shared but do not need a ref count. * + * This attempts to attach to the running transaction in order to account for + * delayed refs, but continues on even when no running transaction exists. + * * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error. */ -int btrfs_check_shared(struct btrfs_trans_handle *trans, - struct btrfs_fs_info *fs_info, u64 root_objectid, - u64 inum, u64 bytenr) +int btrfs_is_data_extent_shared(struct btrfs_inode *inode, u64 bytenr, + u64 extent_gen, + struct btrfs_backref_share_check_ctx *ctx) { - struct ulist *tmp = NULL; - struct ulist *roots = NULL; + struct btrfs_backref_walk_ctx walk_ctx = { 0 }; + struct btrfs_root *root = inode->root; + struct btrfs_fs_info *fs_info = root->fs_info; + struct btrfs_trans_handle *trans; struct ulist_iterator uiter; struct ulist_node *node; - struct seq_list elem = SEQ_LIST_INIT(elem); + struct btrfs_seq_list elem = BTRFS_SEQ_LIST_INIT(elem); int ret = 0; + struct share_check shared = { + .ctx = ctx, + .root = root, + .inum = btrfs_ino(inode), + .data_bytenr = bytenr, + .data_extent_gen = extent_gen, + .share_count = 0, + .self_ref_count = 0, + .have_delayed_delete_refs = false, + }; + int level; + bool leaf_cached; + bool leaf_is_shared; - tmp = ulist_alloc(GFP_NOFS); - roots = ulist_alloc(GFP_NOFS); - if (!tmp || !roots) { - ulist_free(tmp); - ulist_free(roots); - return -ENOMEM; + for (int i = 0; i < BTRFS_BACKREF_CTX_PREV_EXTENTS_SIZE; i++) { + if (ctx->prev_extents_cache[i].bytenr == bytenr) + return ctx->prev_extents_cache[i].is_shared; } - if (trans) - btrfs_get_tree_mod_seq(fs_info, &elem); - else + ulist_init(&ctx->refs); + + trans = btrfs_join_transaction_nostart(root); + if (IS_ERR(trans)) { + if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) { + ret = PTR_ERR(trans); + goto out; + } + trans = NULL; down_read(&fs_info->commit_root_sem); + } else { + btrfs_get_tree_mod_seq(fs_info, &elem); + walk_ctx.time_seq = elem.seq; + } + + ctx->use_path_cache = true; + + /* + * We may have previously determined that the current leaf is shared. + * If it is, then we have a data extent that is shared due to a shared + * subtree (caused by snapshotting) and we don't need to check for data + * backrefs. If the leaf is not shared, then we must do backref walking + * to determine if the data extent is shared through reflinks. + */ + leaf_cached = lookup_backref_shared_cache(ctx, root, + ctx->curr_leaf_bytenr, 0, + &leaf_is_shared); + if (leaf_cached && leaf_is_shared) { + ret = 1; + goto out_trans; + } + + walk_ctx.skip_inode_ref_list = true; + walk_ctx.trans = trans; + walk_ctx.fs_info = fs_info; + walk_ctx.refs = &ctx->refs; + + /* -1 means we are in the bytenr of the data extent. */ + level = -1; ULIST_ITER_INIT(&uiter); while (1) { - ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp, - roots, NULL, root_objectid, inum, 1); - if (ret == BACKREF_FOUND_SHARED) { - /* this is the only condition under which we return 1 */ - ret = 1; + const unsigned long prev_ref_count = ctx->refs.nnodes; + + walk_ctx.bytenr = bytenr; + ret = find_parent_nodes(&walk_ctx, &shared); + if (ret == BACKREF_FOUND_SHARED || + ret == BACKREF_FOUND_NOT_SHARED) { + /* If shared must return 1, otherwise return 0. */ + ret = (ret == BACKREF_FOUND_SHARED) ? 1 : 0; + if (level >= 0) + store_backref_shared_cache(ctx, root, bytenr, + level, ret == 1); break; } if (ret < 0 && ret != -ENOENT) break; ret = 0; - node = ulist_next(tmp, &uiter); + + /* + * More than one extent buffer (bytenr) may have been added to + * the ctx->refs ulist, in which case we have to check multiple + * tree paths in case the first one is not shared, so we can not + * use the path cache which is made for a single path. Multiple + * extent buffers at the current level happen when: + * + * 1) level -1, the data extent: If our data extent was not + * directly shared (without multiple reference items), then + * it might have a single reference item with a count > 1 for + * the same offset, which means there are 2 (or more) file + * extent items that point to the data extent - this happens + * when a file extent item needs to be split and then one + * item gets moved to another leaf due to a b+tree leaf split + * when inserting some item. In this case the file extent + * items may be located in different leaves and therefore + * some of the leaves may be referenced through shared + * subtrees while others are not. Since our extent buffer + * cache only works for a single path (by far the most common + * case and simpler to deal with), we can not use it if we + * have multiple leaves (which implies multiple paths). + * + * 2) level >= 0, a tree node/leaf: We can have a mix of direct + * and indirect references on a b+tree node/leaf, so we have + * to check multiple paths, and the extent buffer (the + * current bytenr) may be shared or not. One example is + * during relocation as we may get a shared tree block ref + * (direct ref) and a non-shared tree block ref (indirect + * ref) for the same node/leaf. + */ + if ((ctx->refs.nnodes - prev_ref_count) > 1) + ctx->use_path_cache = false; + + if (level >= 0) + store_backref_shared_cache(ctx, root, bytenr, + level, false); + node = ulist_next(&ctx->refs, &uiter); if (!node) break; bytenr = node->val; + if (ctx->use_path_cache) { + bool is_shared; + bool cached; + + level++; + cached = lookup_backref_shared_cache(ctx, root, bytenr, + level, &is_shared); + if (cached) { + ret = (is_shared ? 1 : 0); + break; + } + } + shared.share_count = 0; + shared.have_delayed_delete_refs = false; cond_resched(); } - if (trans) + + /* + * If the path cache is disabled, then it means at some tree level we + * got multiple parents due to a mix of direct and indirect backrefs or + * multiple leaves with file extent items pointing to the same data + * extent. We have to invalidate the cache and cache only the sharedness + * result for the levels where we got only one node/reference. + */ + if (!ctx->use_path_cache) { + int i = 0; + + level--; + if (ret >= 0 && level >= 0) { + bytenr = ctx->path_cache_entries[level].bytenr; + ctx->use_path_cache = true; + store_backref_shared_cache(ctx, root, bytenr, level, ret); + i = level + 1; + } + + for ( ; i < BTRFS_MAX_LEVEL; i++) + ctx->path_cache_entries[i].bytenr = 0; + } + + /* + * Cache the sharedness result for the data extent if we know our inode + * has more than 1 file extent item that refers to the data extent. + */ + if (ret >= 0 && shared.self_ref_count > 1) { + int slot = ctx->prev_extents_cache_slot; + + ctx->prev_extents_cache[slot].bytenr = shared.data_bytenr; + ctx->prev_extents_cache[slot].is_shared = (ret == 1); + + slot = (slot + 1) % BTRFS_BACKREF_CTX_PREV_EXTENTS_SIZE; + ctx->prev_extents_cache_slot = slot; + } + +out_trans: + if (trans) { btrfs_put_tree_mod_seq(fs_info, &elem); - else + btrfs_end_transaction(trans); + } else { up_read(&fs_info->commit_root_sem); - ulist_free(tmp); - ulist_free(roots); + } +out: + ulist_release(&ctx->refs); + ctx->prev_leaf_bytenr = ctx->curr_leaf_bytenr; + return ret; } @@ -1649,7 +2048,7 @@ int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid, struct btrfs_key key; struct btrfs_key found_key; struct btrfs_inode_extref *extref; - struct extent_buffer *leaf; + const struct extent_buffer *leaf; unsigned long ptr; key.objectid = inode_objectid; @@ -1733,13 +2132,11 @@ char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, s64 bytes_left = ((s64)size) - 1; struct extent_buffer *eb = eb_in; struct btrfs_key found_key; - int leave_spinning = path->leave_spinning; struct btrfs_inode_ref *iref; if (bytes_left >= 0) dest[bytes_left] = '\0'; - path->leave_spinning = 1; while (1) { bytes_left -= name_len; if (bytes_left >= 0) @@ -1747,7 +2144,7 @@ char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, name_off, name_len); if (eb != eb_in) { if (!path->skip_locking) - btrfs_tree_read_unlock_blocking(eb); + btrfs_tree_read_unlock(eb); free_extent_buffer(eb); } ret = btrfs_find_item(fs_root, path, parent, 0, @@ -1767,8 +2164,6 @@ char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, eb = path->nodes[0]; /* make sure we can use eb after releasing the path */ if (eb != eb_in) { - if (!path->skip_locking) - btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); path->nodes[0] = NULL; path->locks[0] = 0; } @@ -1785,7 +2180,6 @@ char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path, } btrfs_release_path(path); - path->leave_spinning = leave_spinning; if (ret) return ERR_PTR(ret); @@ -1802,26 +2196,33 @@ int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical, struct btrfs_path *path, struct btrfs_key *found_key, u64 *flags_ret) { + struct btrfs_root *extent_root = btrfs_extent_root(fs_info, logical); int ret; u64 flags; u64 size = 0; - u32 item_size; - struct extent_buffer *eb; + const struct extent_buffer *eb; struct btrfs_extent_item *ei; struct btrfs_key key; + key.objectid = logical; if (btrfs_fs_incompat(fs_info, SKINNY_METADATA)) key.type = BTRFS_METADATA_ITEM_KEY; else key.type = BTRFS_EXTENT_ITEM_KEY; - key.objectid = logical; key.offset = (u64)-1; - ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); + ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); if (ret < 0) return ret; + if (unlikely(ret == 0)) { + /* + * Key with offset -1 found, there would have to exist an extent + * item with such offset, but this is out of the valid range. + */ + return -EUCLEAN; + } - ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0); + ret = btrfs_previous_extent_item(extent_root, path, 0); if (ret) { if (ret > 0) ret = -ENOENT; @@ -1841,8 +2242,6 @@ int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical, } eb = path->nodes[0]; - item_size = btrfs_item_size_nr(eb, path->slots[0]); - BUG_ON(item_size < sizeof(*ei)); ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item); flags = btrfs_extent_flags(eb, ei); @@ -1850,7 +2249,7 @@ int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical, btrfs_debug(fs_info, "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u", logical, logical - found_key->objectid, found_key->objectid, - found_key->offset, flags, item_size); + found_key->offset, flags, btrfs_item_size(eb, path->slots[0])); WARN_ON(!flags_ret); if (flags_ret) { @@ -1859,7 +2258,7 @@ int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical, else if (flags & BTRFS_EXTENT_FLAG_DATA) *flags_ret = BTRFS_EXTENT_FLAG_DATA; else - BUG_ON(1); + BUG(); return 0; } @@ -1870,15 +2269,17 @@ int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical, * helper function to iterate extent inline refs. ptr must point to a 0 value * for the first call and may be modified. it is used to track state. * if more refs exist, 0 is returned and the next call to - * __get_extent_inline_ref must pass the modified ptr parameter to get the + * get_extent_inline_ref must pass the modified ptr parameter to get the * next ref. after the last ref was processed, 1 is returned. * returns <0 on error */ -static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb, - struct btrfs_key *key, - struct btrfs_extent_item *ei, u32 item_size, - struct btrfs_extent_inline_ref **out_eiref, - int *out_type) +static int get_extent_inline_ref(unsigned long *ptr, + const struct extent_buffer *eb, + const struct btrfs_key *key, + const struct btrfs_extent_item *ei, + u32 item_size, + struct btrfs_extent_inline_ref **out_eiref, + int *out_type) { unsigned long end; u64 flags; @@ -1908,7 +2309,10 @@ static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb, end = (unsigned long)ei + item_size; *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr); - *out_type = btrfs_extent_inline_ref_type(eb, *out_eiref); + *out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref, + BTRFS_REF_TYPE_ANY); + if (unlikely(*out_type == BTRFS_REF_TYPE_INVALID)) + return -EUCLEAN; *ptr += btrfs_extent_inline_ref_size(*out_type); WARN_ON(*ptr > end); @@ -1921,7 +2325,7 @@ static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb, /* * reads the tree block backref for an extent. tree level and root are returned * through out_level and out_root. ptr must point to a 0 value for the first - * call and may be modified (see __get_extent_inline_ref comment). + * call and may be modified (see get_extent_inline_ref comment). * returns 0 if data was provided, 1 if there was no more data to provide or * <0 on error. */ @@ -1937,7 +2341,7 @@ int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb, return 1; while (1) { - ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size, + ret = get_extent_inline_ref(ptr, eb, key, ei, item_size, &eiref, &type); if (ret < 0) return ret; @@ -1982,7 +2386,7 @@ static int iterate_leaf_refs(struct btrfs_fs_info *fs_info, "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu", extent_item_objectid, eie->inum, eie->offset, root); - ret = iterate(eie->inum, eie->offset, root, ctx); + ret = iterate(eie->inum, eie->offset, eie->num_bytes, root, ctx); if (ret) { btrfs_debug(fs_info, "stopping iteration for %llu due to ret=%d", @@ -1999,104 +2403,181 @@ static int iterate_leaf_refs(struct btrfs_fs_info *fs_info, * the given parameters. * when the iterator function returns a non-zero value, iteration stops. */ -int iterate_extent_inodes(struct btrfs_fs_info *fs_info, - u64 extent_item_objectid, u64 extent_item_pos, - int search_commit_root, - iterate_extent_inodes_t *iterate, void *ctx) +int iterate_extent_inodes(struct btrfs_backref_walk_ctx *ctx, + bool search_commit_root, + iterate_extent_inodes_t *iterate, void *user_ctx) { int ret; - struct btrfs_trans_handle *trans = NULL; - struct ulist *refs = NULL; - struct ulist *roots = NULL; - struct ulist_node *ref_node = NULL; - struct ulist_node *root_node = NULL; - struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem); + struct ulist *refs; + struct ulist_node *ref_node; + struct btrfs_seq_list seq_elem = BTRFS_SEQ_LIST_INIT(seq_elem); struct ulist_iterator ref_uiter; - struct ulist_iterator root_uiter; - btrfs_debug(fs_info, "resolving all inodes for extent %llu", - extent_item_objectid); + btrfs_debug(ctx->fs_info, "resolving all inodes for extent %llu", + ctx->bytenr); + + ASSERT(ctx->trans == NULL); + ASSERT(ctx->roots == NULL); if (!search_commit_root) { - trans = btrfs_join_transaction(fs_info->extent_root); - if (IS_ERR(trans)) - return PTR_ERR(trans); - btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem); + struct btrfs_trans_handle *trans; + + trans = btrfs_attach_transaction(ctx->fs_info->tree_root); + if (IS_ERR(trans)) { + if (PTR_ERR(trans) != -ENOENT && + PTR_ERR(trans) != -EROFS) + return PTR_ERR(trans); + trans = NULL; + } + ctx->trans = trans; + } + + if (ctx->trans) { + btrfs_get_tree_mod_seq(ctx->fs_info, &seq_elem); + ctx->time_seq = seq_elem.seq; } else { - down_read(&fs_info->commit_root_sem); + down_read(&ctx->fs_info->commit_root_sem); } - ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid, - tree_mod_seq_elem.seq, &refs, - &extent_item_pos); + ret = btrfs_find_all_leafs(ctx); if (ret) goto out; + refs = ctx->refs; + ctx->refs = NULL; ULIST_ITER_INIT(&ref_uiter); while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) { - ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val, - tree_mod_seq_elem.seq, &roots); + const u64 leaf_bytenr = ref_node->val; + struct ulist_node *root_node; + struct ulist_iterator root_uiter; + struct extent_inode_elem *inode_list; + + inode_list = (struct extent_inode_elem *)(uintptr_t)ref_node->aux; + + if (ctx->cache_lookup) { + const u64 *root_ids; + int root_count; + bool cached; + + cached = ctx->cache_lookup(leaf_bytenr, ctx->user_ctx, + &root_ids, &root_count); + if (cached) { + for (int i = 0; i < root_count; i++) { + ret = iterate_leaf_refs(ctx->fs_info, + inode_list, + root_ids[i], + leaf_bytenr, + iterate, + user_ctx); + if (ret) + break; + } + continue; + } + } + + if (!ctx->roots) { + ctx->roots = ulist_alloc(GFP_NOFS); + if (!ctx->roots) { + ret = -ENOMEM; + break; + } + } + + ctx->bytenr = leaf_bytenr; + ret = btrfs_find_all_roots_safe(ctx); if (ret) break; + + if (ctx->cache_store) + ctx->cache_store(leaf_bytenr, ctx->roots, ctx->user_ctx); + ULIST_ITER_INIT(&root_uiter); - while (!ret && (root_node = ulist_next(roots, &root_uiter))) { - btrfs_debug(fs_info, + while (!ret && (root_node = ulist_next(ctx->roots, &root_uiter))) { + btrfs_debug(ctx->fs_info, "root %llu references leaf %llu, data list %#llx", root_node->val, ref_node->val, ref_node->aux); - ret = iterate_leaf_refs(fs_info, - (struct extent_inode_elem *) - (uintptr_t)ref_node->aux, - root_node->val, - extent_item_objectid, - iterate, ctx); + ret = iterate_leaf_refs(ctx->fs_info, inode_list, + root_node->val, ctx->bytenr, + iterate, user_ctx); } - ulist_free(roots); + ulist_reinit(ctx->roots); } free_leaf_list(refs); out: - if (!search_commit_root) { - btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem); - btrfs_end_transaction(trans); + if (ctx->trans) { + btrfs_put_tree_mod_seq(ctx->fs_info, &seq_elem); + btrfs_end_transaction(ctx->trans); + ctx->trans = NULL; } else { - up_read(&fs_info->commit_root_sem); + up_read(&ctx->fs_info->commit_root_sem); } + ulist_free(ctx->roots); + ctx->roots = NULL; + + if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP) + ret = 0; + return ret; } +static int build_ino_list(u64 inum, u64 offset, u64 num_bytes, u64 root, void *ctx) +{ + struct btrfs_data_container *inodes = ctx; + const size_t c = 3 * sizeof(u64); + + if (inodes->bytes_left >= c) { + inodes->bytes_left -= c; + inodes->val[inodes->elem_cnt] = inum; + inodes->val[inodes->elem_cnt + 1] = offset; + inodes->val[inodes->elem_cnt + 2] = root; + inodes->elem_cnt += 3; + } else { + inodes->bytes_missing += c - inodes->bytes_left; + inodes->bytes_left = 0; + inodes->elem_missed += 3; + } + + return 0; +} + int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info, - struct btrfs_path *path, - iterate_extent_inodes_t *iterate, void *ctx) + void *ctx, bool ignore_offset) { + struct btrfs_backref_walk_ctx walk_ctx = { 0 }; int ret; - u64 extent_item_pos; u64 flags = 0; struct btrfs_key found_key; - int search_commit_root = path->search_commit_root; + struct btrfs_path *path; + + path = btrfs_alloc_path(); + if (!path) + return -ENOMEM; ret = extent_from_logical(fs_info, logical, path, &found_key, &flags); - btrfs_release_path(path); + btrfs_free_path(path); if (ret < 0) return ret; if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) return -EINVAL; - extent_item_pos = logical - found_key.objectid; - ret = iterate_extent_inodes(fs_info, found_key.objectid, - extent_item_pos, search_commit_root, - iterate, ctx); + walk_ctx.bytenr = found_key.objectid; + if (ignore_offset) + walk_ctx.ignore_extent_item_pos = true; + else + walk_ctx.extent_item_pos = logical - found_key.objectid; + walk_ctx.fs_info = fs_info; - return ret; + return iterate_extent_inodes(&walk_ctx, false, build_ino_list, ctx); } -typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off, - struct extent_buffer *eb, void *ctx); +static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off, + struct extent_buffer *eb, struct inode_fs_paths *ipath); -static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root, - struct btrfs_path *path, - iterate_irefs_t *iterate, void *ctx) +static int iterate_inode_refs(u64 inum, struct inode_fs_paths *ipath) { int ret = 0; int slot; @@ -2105,8 +2586,9 @@ static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root, u32 name_len; u64 parent = 0; int found = 0; + struct btrfs_root *fs_root = ipath->fs_root; + struct btrfs_path *path = ipath->btrfs_path; struct extent_buffer *eb; - struct btrfs_item *item; struct btrfs_inode_ref *iref; struct btrfs_key found_key; @@ -2130,28 +2612,24 @@ static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root, ret = -ENOMEM; break; } - extent_buffer_get(eb); - btrfs_tree_read_lock(eb); - btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); btrfs_release_path(path); - item = btrfs_item_nr(slot); iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref); - for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) { + for (cur = 0; cur < btrfs_item_size(eb, slot); cur += len) { name_len = btrfs_inode_ref_name_len(eb, iref); /* path must be released before calling iterate()! */ btrfs_debug(fs_root->fs_info, "following ref at offset %u for inode %llu in tree %llu", - cur, found_key.objectid, fs_root->objectid); - ret = iterate(parent, name_len, - (unsigned long)(iref + 1), eb, ctx); + cur, found_key.objectid, + btrfs_root_id(fs_root)); + ret = inode_to_path(parent, name_len, + (unsigned long)(iref + 1), eb, ipath); if (ret) break; len = sizeof(*iref) + name_len; iref = (struct btrfs_inode_ref *)((char *)iref + len); } - btrfs_tree_read_unlock_blocking(eb); free_extent_buffer(eb); } @@ -2160,15 +2638,15 @@ static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root, return ret; } -static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root, - struct btrfs_path *path, - iterate_irefs_t *iterate, void *ctx) +static int iterate_inode_extrefs(u64 inum, struct inode_fs_paths *ipath) { int ret; int slot; u64 offset = 0; u64 parent; int found = 0; + struct btrfs_root *fs_root = ipath->fs_root; + struct btrfs_path *path = ipath->btrfs_path; struct extent_buffer *eb; struct btrfs_inode_extref *extref; u32 item_size; @@ -2192,13 +2670,9 @@ static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root, ret = -ENOMEM; break; } - extent_buffer_get(eb); - - btrfs_tree_read_lock(eb); - btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK); btrfs_release_path(path); - item_size = btrfs_item_size_nr(eb, slot); + item_size = btrfs_item_size(eb, slot); ptr = btrfs_item_ptr_offset(eb, slot); cur_offset = 0; @@ -2208,15 +2682,14 @@ static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root, extref = (struct btrfs_inode_extref *)(ptr + cur_offset); parent = btrfs_inode_extref_parent(eb, extref); name_len = btrfs_inode_extref_name_len(eb, extref); - ret = iterate(parent, name_len, - (unsigned long)&extref->name, eb, ctx); + ret = inode_to_path(parent, name_len, + (unsigned long)&extref->name, eb, ipath); if (ret) break; cur_offset += btrfs_inode_extref_name_len(eb, extref); cur_offset += sizeof(*extref); } - btrfs_tree_read_unlock_blocking(eb); free_extent_buffer(eb); offset++; @@ -2227,34 +2700,13 @@ static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root, return ret; } -static int iterate_irefs(u64 inum, struct btrfs_root *fs_root, - struct btrfs_path *path, iterate_irefs_t *iterate, - void *ctx) -{ - int ret; - int found_refs = 0; - - ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx); - if (!ret) - ++found_refs; - else if (ret != -ENOENT) - return ret; - - ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx); - if (ret == -ENOENT && found_refs) - return 0; - - return ret; -} - /* * returns 0 if the path could be dumped (probably truncated) * returns <0 in case of an error */ static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off, - struct extent_buffer *eb, void *ctx) + struct extent_buffer *eb, struct inode_fs_paths *ipath) { - struct inode_fs_paths *ipath = ctx; char *fspath; char *fspath_min; int i = ipath->fspath->elem_cnt; @@ -2295,8 +2747,20 @@ static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off, */ int paths_from_inode(u64 inum, struct inode_fs_paths *ipath) { - return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path, - inode_to_path, ipath); + int ret; + int found_refs = 0; + + ret = iterate_inode_refs(inum, ipath); + if (!ret) + ++found_refs; + else if (ret != -ENOENT) + return ret; + + ret = iterate_inode_extrefs(inum, ipath); + if (ret == -ENOENT && found_refs) + return 0; + + return ret; } struct btrfs_data_container *init_data_container(u32 total_bytes) @@ -2305,20 +2769,14 @@ struct btrfs_data_container *init_data_container(u32 total_bytes) size_t alloc_bytes; alloc_bytes = max_t(size_t, total_bytes, sizeof(*data)); - data = kvmalloc(alloc_bytes, GFP_KERNEL); + data = kvzalloc(alloc_bytes, GFP_KERNEL); if (!data) return ERR_PTR(-ENOMEM); - if (total_bytes >= sizeof(*data)) { + if (total_bytes >= sizeof(*data)) data->bytes_left = total_bytes - sizeof(*data); - data->bytes_missing = 0; - } else { + else data->bytes_missing = sizeof(*data) - total_bytes; - data->bytes_left = 0; - } - - data->elem_cnt = 0; - data->elem_missed = 0; return data; } @@ -2327,7 +2785,7 @@ struct btrfs_data_container *init_data_container(u32 total_bytes) * allocates space to return multiple file system paths for an inode. * total_bytes to allocate are passed, note that space usable for actual path * information will be total_bytes - sizeof(struct inode_fs_paths). - * the returned pointer must be freed with free_ipath() in the end. + * the returned pointer must be freed with __free_inode_fs_paths() in the end. */ struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root, struct btrfs_path *path) @@ -2337,7 +2795,7 @@ struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root, fspath = init_data_container(total_bytes); if (IS_ERR(fspath)) - return (void *)fspath; + return ERR_CAST(fspath); ifp = kmalloc(sizeof(*ifp), GFP_KERNEL); if (!ifp) { @@ -2352,10 +2810,872 @@ struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root, return ifp; } -void free_ipath(struct inode_fs_paths *ipath) +struct btrfs_backref_iter *btrfs_backref_iter_alloc(struct btrfs_fs_info *fs_info) { - if (!ipath) + struct btrfs_backref_iter *ret; + + ret = kzalloc(sizeof(*ret), GFP_NOFS); + if (!ret) + return NULL; + + ret->path = btrfs_alloc_path(); + if (!ret->path) { + kfree(ret); + return NULL; + } + + /* Current backref iterator only supports iteration in commit root */ + ret->path->search_commit_root = true; + ret->path->skip_locking = true; + ret->fs_info = fs_info; + + return ret; +} + +static void btrfs_backref_iter_release(struct btrfs_backref_iter *iter) +{ + iter->bytenr = 0; + iter->item_ptr = 0; + iter->cur_ptr = 0; + iter->end_ptr = 0; + btrfs_release_path(iter->path); + memset(&iter->cur_key, 0, sizeof(iter->cur_key)); +} + +int btrfs_backref_iter_start(struct btrfs_backref_iter *iter, u64 bytenr) +{ + struct btrfs_fs_info *fs_info = iter->fs_info; + struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr); + struct btrfs_path *path = iter->path; + struct btrfs_extent_item *ei; + struct btrfs_key key; + int ret; + + key.objectid = bytenr; + key.type = BTRFS_METADATA_ITEM_KEY; + key.offset = (u64)-1; + iter->bytenr = bytenr; + + ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); + if (ret < 0) + return ret; + if (unlikely(ret == 0)) { + /* + * Key with offset -1 found, there would have to exist an extent + * item with such offset, but this is out of the valid range. + */ + ret = -EUCLEAN; + goto release; + } + if (unlikely(path->slots[0] == 0)) { + DEBUG_WARN(); + ret = -EUCLEAN; + goto release; + } + path->slots[0]--; + + btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); + if ((key.type != BTRFS_EXTENT_ITEM_KEY && + key.type != BTRFS_METADATA_ITEM_KEY) || key.objectid != bytenr) { + ret = -ENOENT; + goto release; + } + memcpy(&iter->cur_key, &key, sizeof(key)); + iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0], + path->slots[0]); + iter->end_ptr = (u32)(iter->item_ptr + + btrfs_item_size(path->nodes[0], path->slots[0])); + ei = btrfs_item_ptr(path->nodes[0], path->slots[0], + struct btrfs_extent_item); + + /* + * Only support iteration on tree backref yet. + * + * This is an extra precaution for non skinny-metadata, where + * EXTENT_ITEM is also used for tree blocks, that we can only use + * extent flags to determine if it's a tree block. + */ + if (btrfs_extent_flags(path->nodes[0], ei) & BTRFS_EXTENT_FLAG_DATA) { + ret = -ENOTSUPP; + goto release; + } + iter->cur_ptr = (u32)(iter->item_ptr + sizeof(*ei)); + + /* If there is no inline backref, go search for keyed backref */ + if (iter->cur_ptr >= iter->end_ptr) { + ret = btrfs_next_item(extent_root, path); + + /* No inline nor keyed ref */ + if (ret > 0) { + ret = -ENOENT; + goto release; + } + if (ret < 0) + goto release; + + btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, + path->slots[0]); + if (iter->cur_key.objectid != bytenr || + (iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY && + iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY)) { + ret = -ENOENT; + goto release; + } + iter->cur_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0], + path->slots[0]); + iter->item_ptr = iter->cur_ptr; + iter->end_ptr = (u32)(iter->item_ptr + btrfs_item_size( + path->nodes[0], path->slots[0])); + } + + return 0; +release: + btrfs_backref_iter_release(iter); + return ret; +} + +static bool btrfs_backref_iter_is_inline_ref(struct btrfs_backref_iter *iter) +{ + if (iter->cur_key.type == BTRFS_EXTENT_ITEM_KEY || + iter->cur_key.type == BTRFS_METADATA_ITEM_KEY) + return true; + return false; +} + +/* + * Go to the next backref item of current bytenr, can be either inlined or + * keyed. + * + * Caller needs to check whether it's inline ref or not by iter->cur_key. + * + * Return 0 if we get next backref without problem. + * Return >0 if there is no extra backref for this bytenr. + * Return <0 if there is something wrong happened. + */ +int btrfs_backref_iter_next(struct btrfs_backref_iter *iter) +{ + struct extent_buffer *eb = iter->path->nodes[0]; + struct btrfs_root *extent_root; + struct btrfs_path *path = iter->path; + struct btrfs_extent_inline_ref *iref; + int ret; + u32 size; + + if (btrfs_backref_iter_is_inline_ref(iter)) { + /* We're still inside the inline refs */ + ASSERT(iter->cur_ptr < iter->end_ptr); + + if (btrfs_backref_has_tree_block_info(iter)) { + /* First tree block info */ + size = sizeof(struct btrfs_tree_block_info); + } else { + /* Use inline ref type to determine the size */ + int type; + + iref = (struct btrfs_extent_inline_ref *) + ((unsigned long)iter->cur_ptr); + type = btrfs_extent_inline_ref_type(eb, iref); + + size = btrfs_extent_inline_ref_size(type); + } + iter->cur_ptr += size; + if (iter->cur_ptr < iter->end_ptr) + return 0; + + /* All inline items iterated, fall through */ + } + + /* We're at keyed items, there is no inline item, go to the next one */ + extent_root = btrfs_extent_root(iter->fs_info, iter->bytenr); + ret = btrfs_next_item(extent_root, iter->path); + if (ret) + return ret; + + btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, path->slots[0]); + if (iter->cur_key.objectid != iter->bytenr || + (iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY && + iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY)) + return 1; + iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0], + path->slots[0]); + iter->cur_ptr = iter->item_ptr; + iter->end_ptr = iter->item_ptr + (u32)btrfs_item_size(path->nodes[0], + path->slots[0]); + return 0; +} + +void btrfs_backref_init_cache(struct btrfs_fs_info *fs_info, + struct btrfs_backref_cache *cache, bool is_reloc) +{ + int i; + + cache->rb_root = RB_ROOT; + for (i = 0; i < BTRFS_MAX_LEVEL; i++) + INIT_LIST_HEAD(&cache->pending[i]); + INIT_LIST_HEAD(&cache->pending_edge); + INIT_LIST_HEAD(&cache->useless_node); + cache->fs_info = fs_info; + cache->is_reloc = is_reloc; +} + +struct btrfs_backref_node *btrfs_backref_alloc_node( + struct btrfs_backref_cache *cache, u64 bytenr, int level) +{ + struct btrfs_backref_node *node; + + ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL); + node = kzalloc(sizeof(*node), GFP_NOFS); + if (!node) + return node; + + INIT_LIST_HEAD(&node->list); + INIT_LIST_HEAD(&node->upper); + INIT_LIST_HEAD(&node->lower); + RB_CLEAR_NODE(&node->rb_node); + cache->nr_nodes++; + node->level = level; + node->bytenr = bytenr; + + return node; +} + +void btrfs_backref_free_node(struct btrfs_backref_cache *cache, + struct btrfs_backref_node *node) +{ + if (node) { + ASSERT(list_empty(&node->list)); + ASSERT(list_empty(&node->lower)); + ASSERT(node->eb == NULL); + cache->nr_nodes--; + btrfs_put_root(node->root); + kfree(node); + } +} + +struct btrfs_backref_edge *btrfs_backref_alloc_edge( + struct btrfs_backref_cache *cache) +{ + struct btrfs_backref_edge *edge; + + edge = kzalloc(sizeof(*edge), GFP_NOFS); + if (edge) + cache->nr_edges++; + return edge; +} + +void btrfs_backref_free_edge(struct btrfs_backref_cache *cache, + struct btrfs_backref_edge *edge) +{ + if (edge) { + cache->nr_edges--; + kfree(edge); + } +} + +void btrfs_backref_unlock_node_buffer(struct btrfs_backref_node *node) +{ + if (node->locked) { + btrfs_tree_unlock(node->eb); + node->locked = 0; + } +} + +void btrfs_backref_drop_node_buffer(struct btrfs_backref_node *node) +{ + if (node->eb) { + btrfs_backref_unlock_node_buffer(node); + free_extent_buffer(node->eb); + node->eb = NULL; + } +} + +/* + * Drop the backref node from cache without cleaning up its children + * edges. + * + * This can only be called on node without parent edges. + * The children edges are still kept as is. + */ +void btrfs_backref_drop_node(struct btrfs_backref_cache *tree, + struct btrfs_backref_node *node) +{ + ASSERT(list_empty(&node->upper)); + + btrfs_backref_drop_node_buffer(node); + list_del_init(&node->list); + list_del_init(&node->lower); + if (!RB_EMPTY_NODE(&node->rb_node)) + rb_erase(&node->rb_node, &tree->rb_root); + btrfs_backref_free_node(tree, node); +} + +/* + * Drop the backref node from cache, also cleaning up all its + * upper edges and any uncached nodes in the path. + * + * This cleanup happens bottom up, thus the node should either + * be the lowest node in the cache or a detached node. + */ +void btrfs_backref_cleanup_node(struct btrfs_backref_cache *cache, + struct btrfs_backref_node *node) +{ + struct btrfs_backref_edge *edge; + + if (!node) return; - kvfree(ipath->fspath); - kfree(ipath); + + while (!list_empty(&node->upper)) { + edge = list_first_entry(&node->upper, struct btrfs_backref_edge, + list[LOWER]); + list_del(&edge->list[LOWER]); + list_del(&edge->list[UPPER]); + btrfs_backref_free_edge(cache, edge); + } + + btrfs_backref_drop_node(cache, node); +} + +/* + * Release all nodes/edges from current cache + */ +void btrfs_backref_release_cache(struct btrfs_backref_cache *cache) +{ + struct btrfs_backref_node *node; + + while ((node = rb_entry_safe(rb_first(&cache->rb_root), + struct btrfs_backref_node, rb_node))) + btrfs_backref_cleanup_node(cache, node); + + ASSERT(list_empty(&cache->pending_edge)); + ASSERT(list_empty(&cache->useless_node)); + ASSERT(!cache->nr_nodes); + ASSERT(!cache->nr_edges); +} + +static void btrfs_backref_link_edge(struct btrfs_backref_edge *edge, + struct btrfs_backref_node *lower, + struct btrfs_backref_node *upper) +{ + ASSERT(upper && lower && upper->level == lower->level + 1); + edge->node[LOWER] = lower; + edge->node[UPPER] = upper; + list_add_tail(&edge->list[LOWER], &lower->upper); +} +/* + * Handle direct tree backref + * + * Direct tree backref means, the backref item shows its parent bytenr + * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined). + * + * @ref_key: The converted backref key. + * For keyed backref, it's the item key. + * For inlined backref, objectid is the bytenr, + * type is btrfs_inline_ref_type, offset is + * btrfs_inline_ref_offset. + */ +static int handle_direct_tree_backref(struct btrfs_backref_cache *cache, + struct btrfs_key *ref_key, + struct btrfs_backref_node *cur) +{ + struct btrfs_backref_edge *edge; + struct btrfs_backref_node *upper; + struct rb_node *rb_node; + + ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY); + + /* Only reloc root uses backref pointing to itself */ + if (ref_key->objectid == ref_key->offset) { + struct btrfs_root *root; + + cur->is_reloc_root = 1; + /* Only reloc backref cache cares about a specific root */ + if (cache->is_reloc) { + root = find_reloc_root(cache->fs_info, cur->bytenr); + if (!root) + return -ENOENT; + cur->root = root; + } else { + /* + * For generic purpose backref cache, reloc root node + * is useless. + */ + list_add(&cur->list, &cache->useless_node); + } + return 0; + } + + edge = btrfs_backref_alloc_edge(cache); + if (!edge) + return -ENOMEM; + + rb_node = rb_simple_search(&cache->rb_root, ref_key->offset); + if (!rb_node) { + /* Parent node not yet cached */ + upper = btrfs_backref_alloc_node(cache, ref_key->offset, + cur->level + 1); + if (!upper) { + btrfs_backref_free_edge(cache, edge); + return -ENOMEM; + } + + /* + * Backrefs for the upper level block isn't cached, add the + * block to pending list + */ + list_add_tail(&edge->list[UPPER], &cache->pending_edge); + } else { + /* Parent node already cached */ + upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node); + ASSERT(upper->checked); + INIT_LIST_HEAD(&edge->list[UPPER]); + } + btrfs_backref_link_edge(edge, cur, upper); + return 0; +} + +/* + * Handle indirect tree backref + * + * Indirect tree backref means, we only know which tree the node belongs to. + * We still need to do a tree search to find out the parents. This is for + * TREE_BLOCK_REF backref (keyed or inlined). + * + * @trans: Transaction handle. + * @ref_key: The same as @ref_key in handle_direct_tree_backref() + * @tree_key: The first key of this tree block. + * @path: A clean (released) path, to avoid allocating path every time + * the function get called. + */ +static int handle_indirect_tree_backref(struct btrfs_trans_handle *trans, + struct btrfs_backref_cache *cache, + struct btrfs_path *path, + struct btrfs_key *ref_key, + struct btrfs_key *tree_key, + struct btrfs_backref_node *cur) +{ + struct btrfs_fs_info *fs_info = cache->fs_info; + struct btrfs_backref_node *upper; + struct btrfs_backref_node *lower; + struct btrfs_backref_edge *edge; + struct extent_buffer *eb; + struct btrfs_root *root; + struct rb_node *rb_node; + int level; + bool need_check = true; + int ret; + + root = btrfs_get_fs_root(fs_info, ref_key->offset, false); + if (IS_ERR(root)) + return PTR_ERR(root); + + /* We shouldn't be using backref cache for non-shareable roots. */ + if (unlikely(!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))) { + btrfs_put_root(root); + return -EUCLEAN; + } + + if (btrfs_root_level(&root->root_item) == cur->level) { + /* Tree root */ + ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr); + /* + * For reloc backref cache, we may ignore reloc root. But for + * general purpose backref cache, we can't rely on + * btrfs_should_ignore_reloc_root() as it may conflict with + * current running relocation and lead to missing root. + * + * For general purpose backref cache, reloc root detection is + * completely relying on direct backref (key->offset is parent + * bytenr), thus only do such check for reloc cache. + */ + if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) { + btrfs_put_root(root); + list_add(&cur->list, &cache->useless_node); + } else { + cur->root = root; + } + return 0; + } + + level = cur->level + 1; + + /* Search the tree to find parent blocks referring to the block */ + path->search_commit_root = true; + path->skip_locking = true; + path->lowest_level = level; + ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0); + path->lowest_level = 0; + if (ret < 0) { + btrfs_put_root(root); + return ret; + } + if (ret > 0 && path->slots[level] > 0) + path->slots[level]--; + + eb = path->nodes[level]; + if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) { + btrfs_err(fs_info, +"couldn't find block (%llu) (level %d) in tree (%llu) with key " BTRFS_KEY_FMT, + cur->bytenr, level - 1, btrfs_root_id(root), + BTRFS_KEY_FMT_VALUE(tree_key)); + btrfs_put_root(root); + ret = -ENOENT; + goto out; + } + lower = cur; + + /* Add all nodes and edges in the path */ + for (; level < BTRFS_MAX_LEVEL; level++) { + if (!path->nodes[level]) { + ASSERT(btrfs_root_bytenr(&root->root_item) == + lower->bytenr); + /* Same as previous should_ignore_reloc_root() call */ + if (btrfs_should_ignore_reloc_root(root) && + cache->is_reloc) { + btrfs_put_root(root); + list_add(&lower->list, &cache->useless_node); + } else { + lower->root = root; + } + break; + } + + edge = btrfs_backref_alloc_edge(cache); + if (!edge) { + btrfs_put_root(root); + ret = -ENOMEM; + goto out; + } + + eb = path->nodes[level]; + rb_node = rb_simple_search(&cache->rb_root, eb->start); + if (!rb_node) { + upper = btrfs_backref_alloc_node(cache, eb->start, + lower->level + 1); + if (!upper) { + btrfs_put_root(root); + btrfs_backref_free_edge(cache, edge); + ret = -ENOMEM; + goto out; + } + upper->owner = btrfs_header_owner(eb); + + /* We shouldn't be using backref cache for non shareable roots. */ + if (unlikely(!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))) { + btrfs_put_root(root); + btrfs_backref_free_edge(cache, edge); + btrfs_backref_free_node(cache, upper); + ret = -EUCLEAN; + goto out; + } + + /* + * If we know the block isn't shared we can avoid + * checking its backrefs. + */ + if (btrfs_block_can_be_shared(trans, root, eb)) + upper->checked = 0; + else + upper->checked = 1; + + /* + * Add the block to pending list if we need to check its + * backrefs, we only do this once while walking up a + * tree as we will catch anything else later on. + */ + if (!upper->checked && need_check) { + need_check = false; + list_add_tail(&edge->list[UPPER], + &cache->pending_edge); + } else { + if (upper->checked) + need_check = true; + INIT_LIST_HEAD(&edge->list[UPPER]); + } + } else { + upper = rb_entry(rb_node, struct btrfs_backref_node, + rb_node); + ASSERT(upper->checked); + INIT_LIST_HEAD(&edge->list[UPPER]); + if (!upper->owner) + upper->owner = btrfs_header_owner(eb); + } + btrfs_backref_link_edge(edge, lower, upper); + + if (rb_node) { + btrfs_put_root(root); + break; + } + lower = upper; + upper = NULL; + } +out: + btrfs_release_path(path); + return ret; +} + +/* + * Add backref node @cur into @cache. + * + * NOTE: Even if the function returned 0, @cur is not yet cached as its upper + * links aren't yet bi-directional. Needs to finish such links. + * Use btrfs_backref_finish_upper_links() to finish such linkage. + * + * @trans: Transaction handle. + * @path: Released path for indirect tree backref lookup + * @iter: Released backref iter for extent tree search + * @node_key: The first key of the tree block + */ +int btrfs_backref_add_tree_node(struct btrfs_trans_handle *trans, + struct btrfs_backref_cache *cache, + struct btrfs_path *path, + struct btrfs_backref_iter *iter, + struct btrfs_key *node_key, + struct btrfs_backref_node *cur) +{ + struct btrfs_backref_edge *edge; + struct btrfs_backref_node *exist; + int ret; + + ret = btrfs_backref_iter_start(iter, cur->bytenr); + if (ret < 0) + return ret; + /* + * We skip the first btrfs_tree_block_info, as we don't use the key + * stored in it, but fetch it from the tree block + */ + if (btrfs_backref_has_tree_block_info(iter)) { + ret = btrfs_backref_iter_next(iter); + if (ret < 0) + goto out; + /* No extra backref? This means the tree block is corrupted */ + if (unlikely(ret > 0)) { + ret = -EUCLEAN; + goto out; + } + } + WARN_ON(cur->checked); + if (!list_empty(&cur->upper)) { + /* + * The backref was added previously when processing backref of + * type BTRFS_TREE_BLOCK_REF_KEY + */ + ASSERT(list_is_singular(&cur->upper)); + edge = list_first_entry(&cur->upper, struct btrfs_backref_edge, + list[LOWER]); + ASSERT(list_empty(&edge->list[UPPER])); + exist = edge->node[UPPER]; + /* + * Add the upper level block to pending list if we need check + * its backrefs + */ + if (!exist->checked) + list_add_tail(&edge->list[UPPER], &cache->pending_edge); + } else { + exist = NULL; + } + + for (; ret == 0; ret = btrfs_backref_iter_next(iter)) { + struct extent_buffer *eb; + struct btrfs_key key; + int type; + + cond_resched(); + eb = iter->path->nodes[0]; + + key.objectid = iter->bytenr; + if (btrfs_backref_iter_is_inline_ref(iter)) { + struct btrfs_extent_inline_ref *iref; + + /* Update key for inline backref */ + iref = (struct btrfs_extent_inline_ref *) + ((unsigned long)iter->cur_ptr); + type = btrfs_get_extent_inline_ref_type(eb, iref, + BTRFS_REF_TYPE_BLOCK); + if (unlikely(type == BTRFS_REF_TYPE_INVALID)) { + ret = -EUCLEAN; + goto out; + } + key.type = type; + key.offset = btrfs_extent_inline_ref_offset(eb, iref); + } else { + key.type = iter->cur_key.type; + key.offset = iter->cur_key.offset; + } + + /* + * Parent node found and matches current inline ref, no need to + * rebuild this node for this inline ref + */ + if (exist && + ((key.type == BTRFS_TREE_BLOCK_REF_KEY && + exist->owner == key.offset) || + (key.type == BTRFS_SHARED_BLOCK_REF_KEY && + exist->bytenr == key.offset))) { + exist = NULL; + continue; + } + + /* SHARED_BLOCK_REF means key.offset is the parent bytenr */ + if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { + ret = handle_direct_tree_backref(cache, &key, cur); + if (ret < 0) + goto out; + } else if (key.type == BTRFS_TREE_BLOCK_REF_KEY) { + /* + * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref + * offset means the root objectid. We need to search + * the tree to get its parent bytenr. + */ + ret = handle_indirect_tree_backref(trans, cache, path, + &key, node_key, cur); + if (ret < 0) + goto out; + } + /* + * Unrecognized tree backref items (if it can pass tree-checker) + * would be ignored. + */ + } + ret = 0; + cur->checked = 1; + WARN_ON(exist); +out: + btrfs_backref_iter_release(iter); + return ret; +} + +/* + * Finish the upwards linkage created by btrfs_backref_add_tree_node() + */ +int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache, + struct btrfs_backref_node *start) +{ + struct list_head *useless_node = &cache->useless_node; + struct btrfs_backref_edge *edge; + struct rb_node *rb_node; + LIST_HEAD(pending_edge); + + ASSERT(start->checked); + + rb_node = rb_simple_insert(&cache->rb_root, &start->simple_node); + if (rb_node) + btrfs_backref_panic(cache->fs_info, start->bytenr, -EEXIST); + + /* + * Use breadth first search to iterate all related edges. + * + * The starting points are all the edges of this node + */ + list_for_each_entry(edge, &start->upper, list[LOWER]) + list_add_tail(&edge->list[UPPER], &pending_edge); + + while (!list_empty(&pending_edge)) { + struct btrfs_backref_node *upper; + struct btrfs_backref_node *lower; + + edge = list_first_entry(&pending_edge, + struct btrfs_backref_edge, list[UPPER]); + list_del_init(&edge->list[UPPER]); + upper = edge->node[UPPER]; + lower = edge->node[LOWER]; + + /* Parent is detached, no need to keep any edges */ + if (upper->detached) { + list_del(&edge->list[LOWER]); + btrfs_backref_free_edge(cache, edge); + + /* Lower node is orphan, queue for cleanup */ + if (list_empty(&lower->upper)) + list_add(&lower->list, useless_node); + continue; + } + + /* + * All new nodes added in current build_backref_tree() haven't + * been linked to the cache rb tree. + * So if we have upper->rb_node populated, this means a cache + * hit. We only need to link the edge, as @upper and all its + * parents have already been linked. + */ + if (!RB_EMPTY_NODE(&upper->rb_node)) { + list_add_tail(&edge->list[UPPER], &upper->lower); + continue; + } + + /* Sanity check, we shouldn't have any unchecked nodes */ + if (unlikely(!upper->checked)) { + DEBUG_WARN("we should not have any unchecked nodes"); + return -EUCLEAN; + } + + rb_node = rb_simple_insert(&cache->rb_root, &upper->simple_node); + if (unlikely(rb_node)) { + btrfs_backref_panic(cache->fs_info, upper->bytenr, -EEXIST); + return -EUCLEAN; + } + + list_add_tail(&edge->list[UPPER], &upper->lower); + + /* + * Also queue all the parent edges of this uncached node + * to finish the upper linkage + */ + list_for_each_entry(edge, &upper->upper, list[LOWER]) + list_add_tail(&edge->list[UPPER], &pending_edge); + } + return 0; +} + +void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache, + struct btrfs_backref_node *node) +{ + struct btrfs_backref_node *lower; + struct btrfs_backref_node *upper; + struct btrfs_backref_edge *edge; + + while (!list_empty(&cache->useless_node)) { + lower = list_first_entry(&cache->useless_node, + struct btrfs_backref_node, list); + list_del_init(&lower->list); + } + while (!list_empty(&cache->pending_edge)) { + edge = list_first_entry(&cache->pending_edge, + struct btrfs_backref_edge, list[UPPER]); + list_del(&edge->list[UPPER]); + list_del(&edge->list[LOWER]); + lower = edge->node[LOWER]; + upper = edge->node[UPPER]; + btrfs_backref_free_edge(cache, edge); + + /* + * Lower is no longer linked to any upper backref nodes and + * isn't in the cache, we can free it ourselves. + */ + if (list_empty(&lower->upper) && + RB_EMPTY_NODE(&lower->rb_node)) + list_add(&lower->list, &cache->useless_node); + + if (!RB_EMPTY_NODE(&upper->rb_node)) + continue; + + /* Add this guy's upper edges to the list to process */ + list_for_each_entry(edge, &upper->upper, list[LOWER]) + list_add_tail(&edge->list[UPPER], + &cache->pending_edge); + if (list_empty(&upper->upper)) + list_add(&upper->list, &cache->useless_node); + } + + while (!list_empty(&cache->useless_node)) { + lower = list_first_entry(&cache->useless_node, + struct btrfs_backref_node, list); + list_del_init(&lower->list); + if (lower == node) + node = NULL; + btrfs_backref_drop_node(cache, lower); + } + + btrfs_backref_cleanup_node(cache, node); + ASSERT(list_empty(&cache->useless_node) && + list_empty(&cache->pending_edge)); } |